save

.specstory/.what-is-this.md

@@ -0,0 +1,65 @@
+
+# SpecStory Artifacts Directory
+    
+This directory is automatically created and maintained by the SpecStory extension to preserve your Cursor composer and chat history.
+    
+## What's Here?
+    
+- `.specstory/history`: Contains markdown files of your AI coding sessions
+    - Each file represents a separate chat or composer session
+    - Files are automatically updated as you work
+- `.specstory/cursor_rules_backups`: Contains backups of the `.cursor/rules/derived-cursor-rules.mdc` file
+    - Backups are automatically created each time the `.cursor/rules/derived-cursor-rules.mdc` file is updated
+    - You can enable/disable the Cursor Rules feature in the SpecStory settings, it is disabled by default
+
+## Valuable Uses
+    
+- Capture: Keep your context window up-to-date when starting new Chat/Composer sessions via @ references
+- Search: For previous prompts and code snippets 
+- Learn: Meta-analyze your patterns and learn from your past experiences
+- Derive: Keep Cursor on course with your past decisions by automatically deriving Cursor rules from your AI interactions
+    
+## Version Control
+    
+We recommend keeping this directory under version control to maintain a history of your AI interactions. However, if you prefer not to version these files, you can exclude them by adding this to your `.gitignore`:
+    
+```
+.specstory
+```
+
+We recommend not keeping the `.specstory/cursor_rules_backups` directory under version control if you are already using git to version the `.cursor/rules` directory, and committing regularly. You can exclude it by adding this to your `.gitignore`:
+
+```
+.specstory/cursor_rules_backups
+```
+
+## Searching Your Codebase
+    
+When searching your codebase in Cursor, search results may include your previous AI coding interactions. To focus solely on your actual code files, you can exclude the AI interaction history from search results.
+    
+To exclude AI interaction history:
+    
+1. Open the "Find in Files" search in Cursor (Cmd/Ctrl + Shift + F)
+2. Navigate to the "files to exclude" section
+3. Add the following pattern:
+    
+```
+.specstory/*
+```
+    
+This will ensure your searches only return results from your working codebase files.
+
+## Notes
+
+- Auto-save only works when Cursor/sqlite flushes data to disk. This results in a small delay after the AI response is complete before SpecStory can save the history.
+- Auto-save does not yet work on remote WSL workspaces.
+
+## Settings
+    
+You can control auto-saving behavior in Cursor:
+    
+1. Open Cursor → Settings → VS Code Settings (Cmd/Ctrl + ,)
+2. Search for "SpecStory"
+3. Find "Auto Save" setting to enable/disable
+    
+Auto-save occurs when changes are detected in Cursor's sqlite database, or every 2 minutes as a safety net.

README.md

@@ -19,6 +19,47 @@ Outil de prétraitement de documents PDF avec agents LLM modulables pour l'analy
 - Tesseract OCR (pour la reconnaissance de texte)
 - Ollama (pour les modèles LLM)
 
+#### Vérification des prérequis (Windows)
+
+Un script de vérification est fourni pour vous aider à identifier les composants manquants:
+
+1. Faites un clic droit sur `check_prerequisites.ps1` et sélectionnez "Exécuter avec PowerShell"
+2. Le script vérifiera tous les prérequis et vous indiquera ce qu'il manque
+3. Suivez les instructions pour installer les composants manquants
+
+### Configuration de l'environnement virtuel
+
+Il est fortement recommandé d'utiliser un environnement virtuel pour installer et exécuter cette application, notamment sur Windows:
+
+#### Avec venv (recommandé)
+
+```bash
+# Windows (PowerShell)
+python -m venv venv
+.\venv\Scripts\Activate.ps1
+
+# Windows (CMD)
+python -m venv venv
+.\venv\Scripts\activate.bat
+
+# Linux/macOS
+python -m venv venv
+source venv/bin/activate
+```
+
+#### Avec Conda (alternative)
+
+```bash
+# Création de l'environnement
+conda create -n ragflow_env python=3.9
+conda activate ragflow_env
+
+# Installation des dépendances via pip
+pip install -r requirements.txt
+```
+
+Une fois l'environnement virtuel activé, vous pouvez procéder à l'installation.
+
 ### Méthode 1 : Installation directe (recommandée)
 
 Clonez ce dépôt et installez avec pip :
@@ -39,24 +80,56 @@ Si vous préférez une installation manuelle :
 pip install -r requirements.txt
 ```
 
+### Méthode 3 : Installation automatisée pour Windows
+
+Pour une installation simplifiée sous Windows, utilisez les scripts d'installation fournis :
+
+#### Avec les scripts batch (.bat)
+
+1. Double-cliquez sur `install_windows.bat`
+2. Le script créera un environnement virtuel et installera toutes les dépendances
+3. Suivez les instructions à l'écran pour les étapes suivantes
+
+Pour lancer l'application par la suite :
+1. Double-cliquez sur `launch_windows.bat`
+2. Le script vérifiera l'installation de Tesseract et Ollama avant de démarrer l'application
+
+#### Avec PowerShell (recommandé pour Windows 10/11)
+
+Si vous préférez utiliser PowerShell (interface plus conviviale avec code couleur) :
+
+1. Faites un clic droit sur `install_windows.ps1` et sélectionnez "Exécuter avec PowerShell"
+2. Suivez les instructions à l'écran pour terminer l'installation
+
+Pour lancer l'application par la suite :
+1. Faites un clic droit sur `launch_windows.ps1` et sélectionnez "Exécuter avec PowerShell"
+2. Le script effectuera des vérifications avant de démarrer l'application
+
+Ces méthodes sont recommandées pour les utilisateurs Windows qui ne sont pas familiers avec les lignes de commande.
+
 ### Installation de Tesseract OCR
 
 Pour l'OCR, vous devez également installer Tesseract :
 
 - **Windows** : Téléchargez et installez depuis [https://github.com/UB-Mannheim/tesseract/wiki](https://github.com/UB-Mannheim/tesseract/wiki)
+  - L'application recherchera Tesseract dans les chemins standards sur Windows (`C:\Program Files\Tesseract-OCR\tesseract.exe`, `C:\Program Files (x86)\Tesseract-OCR\tesseract.exe` ou `C:\Tesseract-OCR\tesseract.exe`)
+  - Assurez-vous d'installer les langues françaises et anglaises pendant l'installation
 - **Linux** : `sudo apt install tesseract-ocr tesseract-ocr-fra tesseract-ocr-eng`
 - **macOS** : `brew install tesseract`
 
-### Installation d'Ollama
+### Connexion au serveur Ollama
 
-Suivez les instructions d'installation d'Ollama disponibles sur [https://ollama.ai/](https://ollama.ai/)
+Cette application se connecte à un serveur Ollama distant pour les fonctionnalités LLM:
 
-Modèles recommandés à télécharger :
-```bash
-ollama pull llava:34b
-ollama pull mistral
-ollama pull llama3.2-vision
-```
+- Adresse du serveur: `217.182.105.173:11434`
+- Modèles utilisés: 
+  - `mistral:latest` (pour le résumé et la traduction légère)
+  - `llava:34b-v1.6-fp16` (pour l'analyse visuelle)
+  - `llama3.2-vision:90b-instruct-q8_0` (pour l'analyse visuelle avancée)
+  - `deepseek-r1:70b-llama-distill-q8_0` (pour le résumé et la traduction avancés)
+  - `qwen2.5:72b-instruct-q8_0` (pour la traduction moyenne)
+
+Assurez-vous que votre machine dispose d'une connexion réseau vers ce serveur. Aucune installation locale d'Ollama n'est nécessaire.
 
 ## Utilisation
 
@@ -89,18 +162,151 @@ python main.py
 L'application propose trois niveaux d'analyse :
 
 | Niveau    | Vision                                | Résumé                        | Traduction                   | Usage recommandé            |
-| --------- | ----------------- | ----------------- | ---------------- | --------------------------- |
-| 🔹 Léger  | llava:34b         | mistral           | mistral          | Débogage, prototypes       |
-| ⚪ Moyen   | llava             | deepseek-r1       | qwen2.5          | Usage normal               |
-| 🔸 Avancé | llama3.2-vision   | deepseek-r1       | deepseek         | Documents critiques        |
+| --------- | ------------------------------------- | ----------------------------- | ---------------------------- | --------------------------- |
+| 🔹 Léger  | llava:34b-v1.6-fp16                  | mistral:latest                | mistral:latest               | Débogage, prototypes       |
+| ⚪ Moyen   | llava:34b-v1.6-fp16                  | deepseek-r1:70b-llama-distill | qwen2.5:72b-instruct        | Usage normal               |
+| 🔸 Avancé | llama3.2-vision:90b-instruct-q8_0    | deepseek-r1:70b-llama-distill | deepseek-r1:70b-llama-distill| Documents critiques        |
 
 Vous pouvez personnaliser ces configurations dans le fichier `config/llm_profiles.json`.
 
-## Paramètres avancés
+## Détail des bibliothèques utilisées
 
-- **Température** : Contrôle la créativité des réponses (0.1-1.0)
-- **Top-p/Top-k** : Paramètres d'échantillonnage pour la génération
-- **Max tokens** : Limite la longueur des réponses générées
+L'application utilise les bibliothèques suivantes, chacune avec un rôle spécifique dans le traitement des documents:
+
+### Bibliothèques principales
+- **PyQt6** : Interface graphique complète (v6.4.0+)
+- **PyMuPDF (fitz)** : Manipulation et rendu des documents PDF (v1.21.0+)
+- **numpy** : Traitement numérique des images et des données (v1.22.0+)
+- **pytesseract** : Interface Python pour Tesseract OCR (v0.3.9+)
+- **Pillow** : Traitement d'images (v9.3.0+)
+- **opencv-python (cv2)** : Traitement d'images avancé et détection de contenu (v4.7.0+)
+- **requests** : Communication avec l'API Ollama (v2.28.0+)
+
+### Rôle de chaque bibliothèque
+- **PyQt6**: Framework d'interface graphique qui gère la visualisation PDF, les sélections interactives, la configuration des agents et l'interface utilisateur globale.
+- **PyMuPDF (fitz)**: Convertit les pages PDF en images, permet d'accéder au contenu des PDF, extraire les pages et obtenir des rendus haute qualité.
+- **numpy**: Manipule les données d'images sous forme de tableaux pour le traitement OCR et la détection de contenu.
+- **pytesseract**: Reconnaît le texte dans les images extraites des PDF, avec support multilingue.
+- **Pillow + opencv-python**: Prétraitement des images avant OCR pour améliorer la reconnaissance du texte.
+- **requests**: Envoie des requêtes au service Ollama local pour utiliser les modèles d'IA.
+
+## Structure du projet et modules clés
+
+```
+ragflow_pretraitement/
+├── main.py                   # Point d'entrée principal
+├── ui/                       # Interface utilisateur
+│   ├── viewer.py             # Visualisation PDF et sélection (PyQt6, fitz)
+│   └── llm_config_panel.py   # Configuration des agents (PyQt6)
+├── agents/                   # Agents LLM
+│   ├── base.py               # Classe de base des agents
+│   ├── vision.py             # Agent d'analyse visuelle (utilise OllamaAPI)
+│   ├── summary.py            # Agent de résumé (utilise OllamaAPI)
+│   ├── translation.py        # Agent de traduction (utilise OllamaAPI)
+│   └── rewriter.py           # Agent de reformulation (utilise OllamaAPI)
+├── utils/                    # Utilitaires
+│   ├── ocr.py                # Reconnaissance de texte (pytesseract, opencv)
+│   ├── markdown_export.py    # Export en Markdown
+│   └── api_ollama.py         # Communication avec l'API Ollama (requests)
+├── config/                   # Configuration
+│   └── llm_profiles.json     # Profils des modèles LLM utilisés
+└── data/                     # Données
+    └── outputs/              # Résultats générés
+```
+
+### Détail des modules principaux
+
+#### Interface utilisateur (ui/)
+- **viewer.py**: Implémente la classe principale `PDFViewer` qui gère la visualisation des PDF, la sélection des zones, le zoom et la navigation entre les pages.
+- **llm_config_panel.py**: Implémente la classe `LLMConfigPanel` qui permet de configurer les agents LLM, leurs paramètres et de lancer les traitements.
+
+#### Agents LLM (agents/)
+- **base.py**: Définit la classe de base `LLMBaseAgent` avec les méthodes communes à tous les agents.
+- **vision.py**: Agent spécialisé dans l'analyse d'images et de schémas.
+- **summary.py**: Agent pour résumer et synthétiser du texte.
+- **translation.py**: Agent pour traduire le contenu (généralement de l'anglais vers le français).
+- **rewriter.py**: Agent pour reformuler et améliorer du texte.
+
+#### Utilitaires (utils/)
+- **ocr.py**: Contient la classe `OCRProcessor` pour extraire du texte des images avec prétraitement avancé.
+- **api_ollama.py**: Implémente la classe `OllamaAPI` pour communiquer avec le service Ollama.
+- **markdown_export.py**: Gère l'export des résultats au format Markdown.
+
+## Suivi des flux de traitement
+
+L'application intègre désormais un système complet de journalisation des flux de traitement entre les différents agents LLM. Cette fonctionnalité permet de suivre et d'analyser en détail chaque étape du processus d'analyse d'images et de texte.
+
+### Flux de traitement d'une image
+
+Pour l'analyse d'une image avec contexte, le flux de traitement complet est le suivant :
+
+1. **Texte en français** → **Agent de traduction** → **Texte en anglais**
+2. **Image + Texte en anglais** → **Agent de vision** → **Analyse en anglais**
+3. **Analyse en anglais** → **Agent de traduction** → **Analyse en français**
+
+Chaque étape est enregistrée avec :
+- Les données d'entrée complètes
+- Les données de sortie complètes
+- Les métadonnées (modèle utilisé, type de contenu, etc.)
+- L'horodatage précis
+
+### Localisation des journaux
+
+Les journaux sont sauvegardés dans les répertoires suivants :
+
+- `data/workflows/` : Journaux du flux de travail complet (format Markdown)
+- `data/translations/` : Journaux détaillés des traductions
+- `data/images/` : Images analysées et résultats des analyses de vision
+- `data/outputs/` : Résultats finaux combinés
+
+### Désactivation des agents
+
+Il est maintenant possible de désactiver certains agents LLM via le fichier de configuration `config/agent_config.py`. Par défaut, l'agent de résumé est désactivé pour permettre l'observation des résultats bruts du flux de traitement.
+
+Pour modifier les agents actifs, vous pouvez éditer le dictionnaire `ACTIVE_AGENTS` dans ce fichier :
+
+```python
+# Configuration des agents activés/désactivés
+ACTIVE_AGENTS = {
+    "ocr": True,         # Agent de reconnaissance optique de caractères
+    "vision": True,      # Agent d'analyse d'images
+    "translation": True, # Agent de traduction
+    "summary": False,    # Agent de résumé (désactivé par défaut)
+    "rewriter": True     # Agent de reformulation
+}
+```
+
+## Adaptations pour Windows
+
+Cette version a été optimisée pour Windows avec les adaptations suivantes:
+
+1. **Chemins Tesseract**: Configuration automatique des chemins Tesseract pour Windows
+2. **URLs API**: Configuration de l'API Ollama pour utiliser `localhost` par défaut
+3. **Chemins de fichiers**: Utilisation de la séparation de chemins compatible Windows
+4. **Compatibilité Unicode**: Support des caractères spéciaux dans les chemins Windows
+
+## Fonctionnalités pratiques et cas d'utilisation
+
+### 1. Analyse de tableaux techniques
+- Sélectionnez un tableau complexe dans un document
+- Utilisez l'agent de vision pour reconnaître la structure
+- Exportez le contenu sous forme de markdown bien formaté
+
+### 2. Traduction de documents techniques
+- Chargez un document en anglais
+- Sélectionnez des sections de texte
+- Utilisez l'agent de traduction pour obtenir une version française précise
+- Exportez le résultat en markdown
+
+### 3. Résumé de documentation longue
+- Identifiez les sections importantes d'un long document
+- Appliquez l'agent de résumé à chaque section
+- Combinez les résultats en un document synthétique
+
+### 4. Analyse de schémas et figures
+- Sélectionnez un schéma complexe
+- Utilisez l'agent de vision pour obtenir une description détaillée
+- Ajoutez du contexte textuel au besoin
 
 ## Résolution des problèmes courants
 
@@ -112,63 +318,35 @@ pip uninstall PyQt6 PyQt6-Qt6 PyQt6-sip
 pip install PyQt6
 ```
 
-### Problèmes avec l'OCR
+### Problèmes avec l'OCR sur Windows
 
 Si l'OCR (Tesseract) ne fonctionne pas correctement :
-1. Vérifiez que Tesseract est correctement installé et disponible dans le PATH
-2. Pour Windows, vous devrez peut-être décommenter et modifier la ligne `pytesseract.pytesseract.tesseract_cmd` dans `utils/ocr.py`
+1. Vérifiez que Tesseract est correctement installé (redémarrez l'application après installation)
+2. Vérifiez l'existence d'un des chemins standards (`C:\Program Files\Tesseract-OCR\tesseract.exe`) 
+3. Si nécessaire, modifiez manuellement le chemin dans `utils/ocr.py`
 
-### Ollama introuvable
+### Connectivité au serveur Ollama
 
-Si Ollama n'est pas détecté, vérifiez que le service est bien démarré :
-```bash
-# Linux/macOS
-ollama serve
+Si vous ne pouvez pas vous connecter au serveur Ollama:
+1. Vérifiez votre connexion réseau et assurez-vous que vous pouvez accéder à `217.182.105.173:11434`
+2. Vérifiez qu'aucun pare-feu ou proxy ne bloque la connexion
+3. Pour vérifier la disponibilité du serveur, essayez d'accéder à `http://217.182.105.173:11434/api/version` dans votre navigateur
 
-# Windows
-# Utilisez l'interface graphique ou exécutez le service Ollama
-```
+Si vous souhaitez utiliser une instance Ollama locale à la place:
+1. Modifiez l'URL dans `utils/api_ollama.py` et `agents/base.py` pour utiliser `http://localhost:11434`
+2. Installez Ollama localement depuis [https://ollama.ai/](https://ollama.ai/)
+3. Téléchargez les modèles requis avec `ollama pull <nom_du_modèle>`
 
-## Structure du projet
+### Modèles manquants
 
-```
-ragflow_pretraitement/
-├── main.py                   # Point d'entrée principal
-├── ui/                       # Interface utilisateur
-│   ├── viewer.py             # Visualisation PDF et sélection
-│   └── llm_config_panel.py   # Configuration des agents
-├── agents/                   # Agents LLM
-│   ├── base.py               # Classe de base
-│   ├── vision.py             # Agent d'analyse visuelle
-│   ├── summary.py            # Agent de résumé
-│   ├── translation.py        # Agent de traduction
-│   └── rewriter.py           # Agent de reformulation
-├── utils/                    # Utilitaires
-│   ├── ocr.py                # Reconnaissance de texte
-│   ├── markdown_export.py    # Export en Markdown
-│   └── api_ollama.py         # Communication avec l'API Ollama
-├── config/                   # Configuration
-│   └── llm_profiles.json     # Profils des LLM
-└── data/                     # Données
-    └── outputs/              # Résultats générés
-```
-
-## Utilisation avancée
-
-### Personnalisation des prompts
-
-Les prompts par défaut peuvent être modifiés directement dans le code des agents (fichiers `agents/*.py`).
-
-### Ajouter de nouveaux modèles
-
-Pour ajouter un nouveau modèle, ajoutez-le dans `config/llm_profiles.json` et assurez-vous qu'il est disponible via Ollama.
+Si certains modèles ne fonctionnent pas, vérifiez leur disponibilité sur le serveur et modifiez si nécessaire les noms dans `config/llm_profiles.json`.
 
 ## Limitations
 
-- OCR parfois imprécis sur les documents complexes
-- Certains modèles nécessitent beaucoup de mémoire GPU
+- OCR parfois imprécis sur les documents complexes ou de mauvaise qualité
+- Certains modèles nécessitent beaucoup de mémoire (8GB+ de RAM recommandé)
 - Les formules mathématiques complexes peuvent être mal interprétées
 
 ## Contribution
 
-Les contributions sont les bienvenues ! N'hésitez pas à soumettre des pull requests ou à signaler des problèmes dans l'outil de suivi. 
+Les contributions sont les bienvenues ! N'hésitez pas à soumettre des pull requests ou à signaler des problèmes. 

agents/base.py

@@ -13,7 +13,7 @@ class LLMBaseAgent:
     pour interagir avec différents modèles de langage.
     """
     
-    def __init__(self, model_name: str, endpoint: str = "http://localhost:11434/v1", **config):
+    def __init__(self, model_name: str, endpoint: str = "http://217.182.105.173:11434", **config):
         """
         Initialise un agent LLM
         

agents/ocr.py

@@ -0,0 +1,185 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+
+"""
+Agent for optical character recognition (OCR) in images
+"""
+
+import os
+import time
+import uuid
+import re
+from typing import Dict, Optional, List, Any, Union
+import pytesseract
+from PIL import Image
+import io
+import platform
+
+from .base import LLMBaseAgent
+
+class OCRAgent(LLMBaseAgent):
+    """
+    Agent for optical character recognition (OCR)
+    """
+    
+    def __init__(self, model_name: str = "ocr", endpoint: str = "", **config):
+        """
+        Initialize the OCR agent
+        
+        Args:
+            model_name (str): Model name (default "ocr" as OCR doesn't use LLM models)
+            endpoint (str): API endpoint (not used for OCR)
+            **config: Additional configuration like language, etc.
+        """
+        # Appeler le constructeur parent avec les paramètres requis
+        super().__init__(model_name, endpoint, **config)
+        
+        # Default configuration for OCR
+        default_config = {
+            "language": "fra",  # Default language: French
+            "tesseract_config": "--psm 1 --oem 3",  # Default Tesseract config
+        }
+        
+        # Merge with provided configuration
+        self.config.update(default_config)
+        for key, value in default_config.items():
+            if key not in self.config:
+                self.config[key] = value
+        
+        # Windows-specific configuration
+        if platform.system() == "Windows":
+            # Possible paths for Tesseract on Windows
+            possible_paths = [
+                r"C:\Program Files\Tesseract-OCR\tesseract.exe",
+                r"C:\Program Files (x86)\Tesseract-OCR\tesseract.exe",
+                r"C:\Tesseract-OCR\tesseract.exe",
+                r"C:\Users\PCDEV\AppData\Local\Programs\Tesseract-OCR\tesseract.exe",
+                r"C:\Users\PCDEV\Tesseract-OCR\tesseract.exe"
+            ]
+            
+            # Look for Tesseract in possible paths
+            tesseract_path = None
+            for path in possible_paths:
+                if os.path.exists(path):
+                    tesseract_path = path
+                    break
+            
+            # Configure pytesseract with the found path
+            if tesseract_path:
+                self.config["tesseract_path"] = tesseract_path
+                pytesseract.pytesseract.tesseract_cmd = tesseract_path
+                print(f"Tesseract found at: {tesseract_path}")
+            else:
+                print("WARNING: Tesseract was not found in standard paths.")
+                print("Please install Tesseract OCR from: https://github.com/UB-Mannheim/tesseract/wiki")
+                print("Or manually specify the path with the tesseract_path parameter")
+
+                # If a path is provided in the configuration, use it anyway
+                if "tesseract_path" in self.config:
+                    pytesseract.pytesseract.tesseract_cmd = self.config["tesseract_path"]
+        
+        # Create directory for OCR logs
+        self.log_dir = os.path.join("data", "ocr_logs")
+        os.makedirs(self.log_dir, exist_ok=True)
+    
+    def generate(self, prompt: str = "", images: Optional[List[bytes]] = None) -> str:
+        """
+        Perform optical character recognition on provided images
+        
+        Args:
+            prompt (str, optional): Not used for OCR
+            images (List[bytes], optional): List of images to process in bytes
+            
+        Returns:
+            str: Text extracted from images
+        """
+        if not images:
+            return "Error: No images provided for OCR"
+        
+        results = []
+        image_count = len(images)
+        
+        # Generate unique ID for this OCR session
+        ocr_id = str(uuid.uuid4())[:8]
+        timestamp = time.strftime("%Y%m%d-%H%M%S")
+        
+        for i, img_bytes in enumerate(images):
+            try:
+                # Open image from bytes
+                img = Image.open(io.BytesIO(img_bytes))
+                
+                # Perform OCR with Tesseract
+                lang = self.config.get("language", "fra")
+                config = self.config.get("tesseract_config", "--psm 1 --oem 3")
+                
+                text = pytesseract.image_to_string(img, lang=lang, config=config)
+                
+                # Basic text cleaning
+                text = self._clean_text(text)
+                
+                if text:
+                    results.append(text)
+                
+                # Save image and OCR result
+                image_path = os.path.join(self.log_dir, f"{timestamp}_{ocr_id}_img{i+1}.png")
+                img.save(image_path, "PNG")
+                
+                # Save extracted text
+                text_path = os.path.join(self.log_dir, f"{timestamp}_{ocr_id}_img{i+1}_ocr.txt")
+                with open(text_path, "w", encoding="utf-8") as f:
+                    f.write(f"OCR Language: {lang}\n")
+                    f.write(f"Tesseract config: {config}\n\n")
+                    f.write(text)
+                
+                print(f"OCR performed on image {i+1}/{image_count}, saved to: {text_path}")
+                
+            except Exception as e:
+                error_msg = f"Error processing image {i+1}: {str(e)}"
+                print(error_msg)
+                
+                # Log the error
+                error_path = os.path.join(self.log_dir, f"{timestamp}_{ocr_id}_img{i+1}_error.txt")
+                with open(error_path, "w", encoding="utf-8") as f:
+                    f.write(f"Error processing image {i+1}:\n{str(e)}")
+                
+                # Add error message to results
+                results.append(f"[OCR Error on image {i+1}: {str(e)}]")
+        
+        # Combine all extracted texts
+        if not results:
+            return "No text could be extracted from the provided images."
+        
+        combined_result = "\n\n".join(results)
+        
+        # Save combined result
+        combined_path = os.path.join(self.log_dir, f"{timestamp}_{ocr_id}_combined.txt")
+        with open(combined_path, "w", encoding="utf-8") as f:
+            f.write(f"OCR Language: {self.config.get('language', 'fra')}\n")
+            f.write(f"Number of images: {image_count}\n\n")
+            f.write(combined_result)
+        
+        return combined_result
+    
+    def _clean_text(self, text: str) -> str:
+        """
+        Clean the text extracted by OCR
+        
+        Args:
+            text (str): Raw text to clean
+            
+        Returns:
+            str: Cleaned text
+        """
+        if not text:
+            return ""
+        
+        # Remove spaces at beginning and end
+        text = text.strip()
+        
+        # Remove multiple empty lines
+        text = re.sub(r'\n{3,}', '\n\n', text)
+        
+        # Remove non-printable characters
+        text = ''.join(c for c in text if c.isprintable() or c == '\n')
+        
+        return text 

agents/rewriter.py

@@ -5,25 +5,29 @@
 Agent LLM pour la reformulation et l'adaptation de contenu
 """
 
-import requests
-from typing import Dict, Optional, Union
+import os
+import time
+import uuid
+from typing import Dict, Optional, Union, List, Any
 
 from .base import LLMBaseAgent
+from utils.api_ollama import OllamaAPI
 
 class RewriterAgent(LLMBaseAgent):
     """
     Agent LLM spécialisé dans la reformulation et l'adaptation de texte
     """
     
-    def __init__(self, model_name: str = "mistral", **config):
+    def __init__(self, model_name: str, endpoint: str = "http://217.182.105.173:11434", **config):
         """
         Initialise l'agent de reformulation
         
         Args:
-            model_name (str): Nom du modèle de reformulation (défaut: mistral)
+            model_name (str): Nom du modèle de reformulation
+            endpoint (str): URL de l'API Ollama
             **config: Paramètres de configuration supplémentaires
         """
-        super().__init__(model_name, **config)
+        super().__init__(model_name, endpoint, **config)
         
         # Définir les modes de reformulation et leurs prompts
         self.modes = {
@@ -83,26 +87,38 @@ class RewriterAgent(LLMBaseAgent):
             }
         }
         
-    def generate(self, text: str, mode: str = "rag", custom_prompt: Optional[str] = "") -> str:
+        # Création du répertoire pour les journaux de reformulation
+        self.log_dir = os.path.join("data", "rewrites")
+        os.makedirs(self.log_dir, exist_ok=True)
+    
+    def generate(self, text: str, images: Optional[List[bytes]] = None, mode: str = "rag", 
+                 custom_prompt: Optional[str] = "", language: Optional[str] = None) -> str:
         """
         Reformule un texte selon le mode spécifié
         
         Args:
             text (str): Texte à reformuler
+            images (List[bytes], optional): Non utilisé pour la reformulation
             mode (str): Mode de reformulation (simplifier, détailler, rag, formal, bullet)
             custom_prompt (str, optional): Prompt personnalisé pour la reformulation
+            language (str, optional): Langue du texte à reformuler
             
         Returns:
             str: Le texte reformulé
         """
-        if not text:
-            return ""
+        if not text or not text.strip():
+            return "Erreur: Aucun texte fourni pour la reformulation"
         
         # Déterminer la langue et le prompt à utiliser
-        lang = self.config.get("language", "fr")
+        lang = language or self.config.get("language", "fr")
+        
+        # Génération d'un ID unique pour cette reformulation
+        rewrite_id = str(uuid.uuid4())[:8]
+        timestamp = time.strftime("%Y%m%d-%H%M%S")
         
         if custom_prompt:
-            prompt = custom_prompt.format(text=text)
+            prompt_template = custom_prompt
+            mode_name = "custom"
         else:
             # Vérifier que le mode existe pour la langue spécifiée
             if lang not in self.modes or mode not in self.modes[lang]:
@@ -111,67 +127,81 @@ class RewriterAgent(LLMBaseAgent):
                 mode = "rag"
             
             prompt_template = self.modes[lang][mode]
+            mode_name = mode
+        
+        # Formater le prompt avec le texte
         prompt = prompt_template.format(text=text)
         
+        # Créer l'API Ollama pour l'appel direct
+        api = OllamaAPI(base_url=self.endpoint)
+        
+        # Journaliser le prompt complet
+        print(f"Envoi du prompt de reformulation (mode {mode_name}) au modèle {self.model_name}")
+        
         try:
-            # Construire la payload pour l'API Ollama
-            payload = {
-                "model": self.model_name,
-                "prompt": prompt,
-                "options": {
-                    "temperature": self.config.get("temperature", 0.3),  # Légèrement plus créatif pour la reformulation
+            # Pour les modèles qui supportent le format de chat
+            if any(name in self.model_name.lower() for name in ["llama", "mistral", "deepseek", "qwen"]):
+                # Formater en tant que messages de chat
+                system_prompt = f"Tu es un expert en reformulation de texte spécialisé dans le mode '{mode_name}'."
+                
+                messages = [
+                    {"role": "system", "content": system_prompt},
+                    {"role": "user", "content": prompt}
+                ]
+                
+                response = api.chat(
+                    model=self.model_name,
+                    messages=messages,
+                    options={
+                        "temperature": self.config.get("temperature", 0.3),
                         "top_p": self.config.get("top_p", 0.95),
                         "top_k": self.config.get("top_k", 40),
                         "num_predict": self.config.get("max_tokens", 2048)
                     }
+                )
+                
+                if "message" in response and "content" in response["message"]:
+                    result = response["message"]["content"]
+                else:
+                    result = response.get("response", "Erreur: Format de réponse inattendu")
+            else:
+                # Format de génération standard pour les autres modèles
+                response = api.generate(
+                    model=self.model_name,
+                    prompt=prompt,
+                    options={
+                        "temperature": self.config.get("temperature", 0.3),
+                        "top_p": self.config.get("top_p", 0.95),
+                        "top_k": self.config.get("top_k", 40),
+                        "num_predict": self.config.get("max_tokens", 2048)
                     }
+                )
                 
-            # Dans une implémentation réelle, envoyer la requête à l'API Ollama
-            # response = requests.post(f"{self.endpoint}/api/generate", json=payload)
-            # json_response = response.json()
-            # return json_response.get("response", "")
+                result = response.get("response", "Erreur: Pas de réponse")
             
-            # Pour cette démonstration, retourner des exemples de reformulation
-            if mode == "simplifier" or mode == "simplify":
-                return ("Ce texte a été simplifié pour être plus facile à comprendre. "
-                        "Les mots compliqués ont été remplacés par des mots plus simples. "
-                        "Les phrases longues ont été raccourcies. Les idées principales "
-                        "restent les mêmes, mais elles sont expliquées plus clairement.")
+            # Enregistrer la reformulation dans un fichier
+            rewrite_path = os.path.join(self.log_dir, f"{timestamp}_{rewrite_id}.txt")
+            with open(rewrite_path, "w", encoding="utf-8") as f:
+                f.write(f"Language: {lang}\n")
+                f.write(f"Mode: {mode_name}\n")
+                f.write(f"Model: {self.model_name}\n\n")
+                f.write(f"Original text:\n{text}\n\n")
+                f.write(f"Rewritten text:\n{result}")
             
-            elif mode == "détailler" or mode == "elaborate":
-                return (f"{text}\n\nEn outre, il est important de noter que ce contenu s'inscrit "
-                        "dans un contexte plus large. Plusieurs exemples concrets illustrent ce point : "
-                        "premièrement, l'application pratique de ces concepts dans des situations réelles; "
-                        "deuxièmement, les différentes interprétations possibles selon le domaine d'expertise; "
-                        "et troisièmement, les implications à long terme de ces informations. "
-                        "Cette perspective élargie permet une compréhension plus approfondie du sujet.")
-            
-            elif mode == "formal":
-                return ("Il convient de préciser que le contenu susmentionné présente des caractéristiques "
-                        "particulièrement pertinentes dans le cadre de l'analyse proposée. En effet, "
-                        "l'examen minutieux des éléments constitutifs révèle une structure cohérente "
-                        "dont la logique sous-jacente manifeste une organisation méthodique des concepts. "
-                        "Par conséquent, il est possible d'affirmer que les principes énoncés "
-                        "s'inscrivent dans un paradigme rigoureux qui mérite une attention scientifique.")
-            
-            elif mode == "bullet":
-                # Transformation en liste à puces
-                bullet_points = ["• Point principal: Le contenu présente des informations essentielles",
-                                 "• Structure: Organisation logique des éléments clés",
-                                 "• Application: Utilisations pratiques dans divers contextes",
-                                 "• Avantages: Amélioration de la compréhension et de l'efficacité",
-                                 "• Limitations: Considérations importantes à prendre en compte"]
-                
-                return "\n".join(bullet_points)
-            
-            else:  # rag par défaut
-                return ("Ce contenu optimisé pour les systèmes RAG contient des mots-clés pertinents "
-                        "et une structure sémantique améliorée. Les concepts principaux sont clairement "
-                        "définis et leurs relations sont explicitement établies. Les informations "
-                        "sont présentées de manière à faciliter la recherche et la récupération "
-                        "automatisées, avec une organisation logique qui met en évidence les "
-                        "éléments essentiels du sujet traité. Chaque section est conçue pour "
-                        "maximiser la pertinence lors des requêtes d'information.")
+            print(f"Reformulation enregistrée dans: {rewrite_path}")
+            return result
             
         except Exception as e:
-            return f"Erreur lors de la reformulation: {str(e)}" 
+            error_msg = f"Erreur lors de la reformulation: {str(e)}"
+            print(error_msg)
+            
+            # Enregistrer l'erreur
+            error_path = os.path.join(self.log_dir, f"{timestamp}_{rewrite_id}_error.txt")
+            with open(error_path, "w", encoding="utf-8") as f:
+                f.write(f"Language: {lang}\n")
+                f.write(f"Mode: {mode_name}\n")
+                f.write(f"Model: {self.model_name}\n\n")
+                f.write(f"Original text:\n{text}\n\n")
+                f.write(f"Error:\n{str(e)}")
+                
+            return error_msg 

agents/summary.py

@@ -2,182 +2,175 @@
 # -*- coding: utf-8 -*-
 
 """
-Agent LLM pour la génération de résumés
+Agent pour la synthèse et le résumé de texte
 """
 
-import requests
-from typing import Dict, Optional, Union
+import os
+import time
+import uuid
+from typing import List, Optional, Dict, Any
 
 from .base import LLMBaseAgent
+from utils.api_ollama import OllamaAPI
 
 class SummaryAgent(LLMBaseAgent):
     """
-    Agent LLM spécialisé dans la génération de résumés et d'analyses de texte
+    Agent pour la synthèse et le résumé de texte
     """
     
-    def __init__(self, model_name: str = "deepseek-r1", **config):
+    def __init__(self, model_name: str, endpoint: str = "http://217.182.105.173:11434", **config):
         """
         Initialise l'agent de résumé
         
         Args:
-            model_name (str): Nom du modèle de résumé (défaut: deepseek-r1)
-            **config: Paramètres de configuration supplémentaires
+            model_name (str): Nom du modèle à utiliser
+            endpoint (str): URL de l'API Ollama
+            **config: Configuration supplémentaire
         """
-        super().__init__(model_name, **config)
+        # Appeler le constructeur de la classe parent avec les paramètres requis
+        super().__init__(model_name, endpoint, **config)
         
-        # Définir les prompts de résumé par type de contenu et langue
-        self.prompts = {
-            "fr": {
-                "standard": "Résume le texte suivant en capturant les points essentiels "
-                            "de manière concise et précise. Préserve les informations clés "
-                            "tout en réduisant la longueur.\n\n"
-                            "Texte à résumer :\n{text}",
-                
-                "analytique": "Analyse le texte suivant en identifiant les concepts clés, "
-                              "les arguments principaux et les implications. Fournis un résumé "
-                              "qui met en évidence les insights importants.\n\n"
-                              "Texte à analyser :\n{text}",
-                
-                "schéma": "Décris ce schéma de manière concise en expliquant sa structure, "
-                          "son organisation et les relations entre ses éléments. Identifie "
-                          "le message principal qu'il communique.\n\n"
-                          "Description du schéma :\n{text}",
-                
-                "tableau": "Résume les informations essentielles présentées dans ce tableau "
-                           "en identifiant les tendances, les valeurs importantes et les "
-                           "relations entre les données.\n\n"
-                           "Description du tableau :\n{text}",
-                
-                "formule": "Explique cette formule mathématique de manière concise et accessible, "
-                           "en précisant son contexte, sa signification et ses applications "
-                           "pratiques.\n\n"
-                           "Description de la formule :\n{text}"
-            },
-            "en": {
-                "standard": "Summarize the following text by capturing the essential points "
-                            "in a concise and accurate manner. Preserve key information "
-                            "while reducing length.\n\n"
-                            "Text to summarize:\n{text}",
-                
-                "analytical": "Analyze the following text by identifying key concepts, "
-                              "main arguments, and implications. Provide a summary "
-                              "that highlights important insights.\n\n"
-                              "Text to analyze:\n{text}",
-                
-                "schéma": "Describe this diagram concisely by explaining its structure, "
-                          "organization, and relationships between elements. Identify "
-                          "the main message it communicates.\n\n"
-                          "Diagram description:\n{text}",
-                
-                "tableau": "Summarize the essential information presented in this table "
-                           "by identifying trends, important values, and relationships "
-                           "between the data.\n\n"
-                           "Table description:\n{text}",
-                
-                "formule": "Explain this mathematical formula concisely and accessibly, "
-                           "specifying its context, meaning, and practical applications.\n\n"
-                           "Formula description:\n{text}"
-            }
+        # Configuration par défaut pour les résumés
+        default_config = {
+            "language": "fr",
+            "summary_length": "medium"  # 'short', 'medium', 'long'
         }
         
-    def generate(self, text: str, summary_type: str = "standard", selection_type: Optional[str] = "") -> str:
+        # Mettre à jour la configuration avec les valeurs par défaut si elles ne sont pas spécifiées
+        for key, value in default_config.items():
+            if key not in self.config:
+                self.config[key] = value
+        
+        # Création du répertoire pour les journaux de résumé
+        self.log_dir = os.path.join("data", "summaries")
+        os.makedirs(self.log_dir, exist_ok=True)
+    
+    def generate(self, prompt: str, images: Optional[List[bytes]] = None, 
+                 summary_length: Optional[str] = None, language: Optional[str] = None) -> str:
         """
-        Génère un résumé ou une analyse du texte fourni
+        Résume un texte
         
         Args:
-            text (str): Texte à résumer ou analyser
-            summary_type (str): Type de résumé ("standard" ou "analytique")
-            selection_type (str, optional): Type de contenu ("schéma", "tableau", "formule")
-                Si fourni, remplace summary_type
+            prompt (str): Texte à résumer
+            images (List[bytes], optional): Non utilisé pour le résumé
+            summary_length (str, optional): Longueur du résumé ('short', 'medium', 'long')
+            language (str, optional): Langue du résumé
             
         Returns:
-            str: Le résumé ou l'analyse générée par le modèle
+            str: Résumé généré
         """
-        if not text:
-            return ""
+        if not prompt or not prompt.strip():
+            return "Erreur: Aucun texte fourni pour le résumé"
         
-        # Déterminer le type de prompt à utiliser
-        lang = self.config.get("language", "fr")
+        # Utiliser les paramètres spécifiés ou ceux de la configuration
+        length = summary_length or self.config["summary_length"]
+        lang = language or self.config["language"]
         
-        # Utiliser selection_type seulement s'il est non vide et existe dans les prompts
-        if selection_type and selection_type in self.prompts.get(lang, {}):
-            prompt_type = selection_type
+        # Génération d'un ID unique pour ce résumé
+        summary_id = str(uuid.uuid4())[:8]
+        timestamp = time.strftime("%Y%m%d-%H%M%S")
+        
+        # Définir les instructions de longueur en fonction de la langue
+        length_instructions = {
+            "short": {
+                "fr": "Fais un résumé court et concis (environ 2-3 phrases).",
+                "en": "Create a short and concise summary (about 2-3 sentences)."
+            },
+            "medium": {
+                "fr": "Fais un résumé de taille moyenne (environ 1-2 paragraphes).",
+                "en": "Create a medium-length summary (about 1-2 paragraphs)."
+            },
+            "long": {
+                "fr": "Fais un résumé détaillé (environ 3-4 paragraphes).",
+                "en": "Create a detailed summary (about 3-4 paragraphs)."
+            }
+        }
+        
+        # Obtenir l'instruction de longueur dans la langue appropriée
+        length_instruction = length_instructions.get(length, length_instructions["medium"]).get(lang, length_instructions["medium"]["fr"])
+        
+        # Construire le prompt pour le modèle
+        if lang == "fr":
+            system_prompt = f"Tu es un expert en synthèse et résumé de textes. {length_instruction} "
+            system_prompt += "Conserve l'information essentielle et la structure logique du texte original."
         else:
-            prompt_type = summary_type
+            system_prompt = f"You are an expert in summarizing texts. {length_instruction} "
+            system_prompt += "Preserve the essential information and logical structure of the original text."
         
-        # Obtenir le prompt approprié
-        if lang not in self.prompts or prompt_type not in self.prompts[lang]:
-            lang = "fr"  # Langue par défaut
-            prompt_type = "standard"  # Type par défaut
+        # Construire le message utilisateur avec le texte à résumer
+        user_prompt = prompt
         
-        prompt_template = self.prompts[lang][prompt_type]
-        prompt = prompt_template.format(text=text)
+        # Créer l'API Ollama pour l'appel direct
+        api = OllamaAPI(base_url=self.endpoint)
+        
+        # Journaliser le prompt complet
+        full_prompt = f"System: {system_prompt}\n\nUser: {user_prompt}"
+        print(f"Envoi du prompt de résumé au modèle {self.model_name}:\n{system_prompt}")
         
         try:
-            # Construire la payload pour l'API Ollama
-            payload = {
-                "model": self.model_name,
-                "prompt": prompt,
-                "options": {
+            # Pour les modèles qui supportent le format de chat
+            if any(name in self.model_name.lower() for name in ["llama", "mistral", "deepseek", "qwen"]):
+                # Formater en tant que messages de chat
+                messages = [
+                    {"role": "system", "content": system_prompt},
+                    {"role": "user", "content": user_prompt}
+                ]
+                
+                response = api.chat(
+                    model=self.model_name,
+                    messages=messages,
+                    options={
                         "temperature": self.config.get("temperature", 0.2),
                         "top_p": self.config.get("top_p", 0.95),
                         "top_k": self.config.get("top_k", 40),
-                    "num_predict": self.config.get("max_tokens", 1024)
+                        "num_predict": self.config.get("max_tokens", 2048)
                     }
+                )
+                
+                if "message" in response and "content" in response["message"]:
+                    result = response["message"]["content"]
+                else:
+                    result = response.get("response", "Erreur: Format de réponse inattendu")
+            else:
+                # Format de génération standard pour les autres modèles
+                prompt_text = f"{system_prompt}\n\n{user_prompt}"
+                
+                response = api.generate(
+                    model=self.model_name,
+                    prompt=prompt_text,
+                    options={
+                        "temperature": self.config.get("temperature", 0.2),
+                        "top_p": self.config.get("top_p", 0.95),
+                        "top_k": self.config.get("top_k", 40),
+                        "num_predict": self.config.get("max_tokens", 2048)
                     }
+                )
                 
-            # Dans une implémentation réelle, envoyer la requête à l'API Ollama
-            # response = requests.post(f"{self.endpoint}/api/generate", json=payload)
-            # json_response = response.json()
-            # return json_response.get("response", "")
+                result = response.get("response", "Erreur: Pas de réponse")
             
-            # Pour cette démonstration, retourner un résumé simulé
-            length = len(text.split())
+            # Enregistrer le résumé dans un fichier
+            summary_path = os.path.join(self.log_dir, f"{timestamp}_{summary_id}.txt")
+            with open(summary_path, "w", encoding="utf-8") as f:
+                f.write(f"Language: {lang}\n")
+                f.write(f"Summary length: {length}\n")
+                f.write(f"Model: {self.model_name}\n\n")
+                f.write(f"Original text:\n{prompt}\n\n")
+                f.write(f"Summary:\n{result}")
             
-            if prompt_type == "schéma":
-                return ("Ce schéma illustre une structure organisée présentant les relations "
-                        "entre différents éléments clés. Il met en évidence la hiérarchie et "
-                        "le flux d'information, permettant de comprendre rapidement le processus "
-                        "ou le système représenté. Les composants principaux sont clairement "
-                        "identifiés et leurs connections logiques sont bien établies.")
-            
-            elif prompt_type == "tableau":
-                return ("Ce tableau présente des données structurées qui révèlent des tendances "
-                        "significatives. Les valeurs clés montrent une corrélation entre les "
-                        "différentes variables présentées. L'organisation des données permet "
-                        "d'identifier rapidement les informations importantes et de comprendre "
-                        "les relations entre elles. Les catégories principales sont clairement "
-                        "distinguées.")
-            
-            elif prompt_type == "formule":
-                return ("Cette formule mathématique exprime une relation fondamentale entre "
-                        "plusieurs variables. Elle permet de calculer précisément les valeurs "
-                        "recherchées en fonction des paramètres d'entrée. Son application "
-                        "pratique concerne principalement la modélisation du phénomène décrit "
-                        "dans le contexte. La structure de l'équation révèle la nature des "
-                        "interactions entre les différentes composantes du système.")
-            
-            elif prompt_type == "analytique":
-                return ("L'analyse approfondie du texte révèle plusieurs concepts clés interconnectés. "
-                        "Les arguments principaux s'articulent autour d'une thèse centrale qui est "
-                        "étayée par des preuves pertinentes. Les implications de ces idées sont "
-                        "significatives pour le domaine concerné. Cette analyse met en lumière "
-                        "les points essentiels tout en préservant la profondeur intellectuelle "
-                        "du contenu original.")
-            
-            else:  # standard
-                # Génération d'un résumé approximativement 70% plus court
-                words = text.split()
-                summary_length = max(3, int(length * 0.3))  # Au moins 3 mots
-                
-                if length < 20:
-                    return text  # Texte déjà court
-                
-                return ("Ce texte présente de manière concise les informations essentielles "
-                        "du contenu original. Les points clés sont préservés tout en éliminant "
-                        "les détails superflus. La synthèse capture l'essence du message "
-                        "en mettant l'accent sur les éléments les plus pertinents pour "
-                        "la compréhension globale du sujet traité.")
+            print(f"Résumé enregistré dans: {summary_path}")
+            return result
             
         except Exception as e:
-            return f"Erreur lors de la génération du résumé: {str(e)}" 
+            error_msg = f"Erreur lors de la génération du résumé: {str(e)}"
+            print(error_msg)
+            
+            # Enregistrer l'erreur
+            error_path = os.path.join(self.log_dir, f"{timestamp}_{summary_id}_error.txt")
+            with open(error_path, "w", encoding="utf-8") as f:
+                f.write(f"Language: {lang}\n")
+                f.write(f"Summary length: {length}\n")
+                f.write(f"Model: {self.model_name}\n\n")
+                f.write(f"Original text:\n{prompt}\n\n")
+                f.write(f"Error:\n{str(e)}")
+                
+            return error_msg 

agents/translation.py

@@ -2,140 +2,151 @@
 # -*- coding: utf-8 -*-
 
 """
-Agent LLM pour la traduction de contenu
+Agent pour la traduction de texte
 """
 
-import json
-import requests
-from typing import Optional, Dict
+import os
+import time
+import uuid
+from typing import List, Optional, Dict, Any
 
 from .base import LLMBaseAgent
+from utils.api_ollama import OllamaAPI
 
 class TranslationAgent(LLMBaseAgent):
     """
-    Agent LLM spécialisé dans la traduction de texte
+    Agent pour la traduction de texte
     """
     
-    def __init__(self, model_name: str = "mistral", **config):
+    def __init__(self, model_name: str, endpoint: str = "http://217.182.105.173:11434", **config):
         """
         Initialise l'agent de traduction
         
         Args:
-            model_name (str): Nom du modèle de traduction (défaut: mistral)
-            **config: Paramètres de configuration supplémentaires
+            model_name (str): Nom du modèle à utiliser
+            endpoint (str): URL de l'API Ollama
+            **config: Configuration supplémentaire
         """
-        super().__init__(model_name, **config)
+        super().__init__(model_name, endpoint, **config)
         
-        # Définir les prompts de traduction
-        self.prompts = {
-            "fr_to_en": "Traduis le texte suivant du français vers l'anglais. "
-                        "Préserve le formatage, le ton et le style du texte original. "
-                        "Assure-toi que la traduction est fluide et naturelle.\n\n"
-                        "Texte français :\n{text}",
-            
-            "en_to_fr": "Traduis le texte suivant de l'anglais vers le français. "
-                        "Préserve le formatage, le ton et le style du texte original. "
-                        "Assure-toi que la traduction est fluide et naturelle.\n\n"
-                        "Texte anglais :\n{text}"
+        # Configuration par défaut pour la traduction
+        default_config = {
+            "source_language": "en",
+            "target_language": "fr"
         }
         
-    def generate(self, text: str, source_lang: str = "fr", target_lang: str = "en") -> str:
+        # Mettre à jour la configuration avec les valeurs par défaut si elles ne sont pas spécifiées
+        for key, value in default_config.items():
+            if key not in self.config:
+                self.config[key] = value
+        
+        # Création du répertoire pour les journaux de traduction
+        self.log_dir = os.path.join("data", "translations")
+        os.makedirs(self.log_dir, exist_ok=True)
+    
+    def generate(self, prompt: str, images: Optional[List[bytes]] = None, 
+                 source_language: Optional[str] = None, target_language: Optional[str] = None) -> str:
         """
-        Traduit un texte d'une langue source vers une langue cible
+        Traduit un texte
         
         Args:
-            text (str): Texte à traduire
-            source_lang (str): Langue source (fr ou en)
-            target_lang (str): Langue cible (fr ou en)
+            prompt (str): Texte à traduire
+            images (List[bytes], optional): Non utilisé pour la traduction
+            source_language (str, optional): Langue source (par défaut: celle de la configuration)
+            target_language (str, optional): Langue cible (par défaut: celle de la configuration)
             
         Returns:
-            str: La traduction générée par le modèle
+            str: Traduction générée
         """
-        if not text:
-            return ""
+        if not prompt or not prompt.strip():
+            return "Erreur: Aucun texte fourni pour la traduction"
         
-        # Déterminer la direction de traduction
-        if source_lang == "fr" and target_lang == "en":
-            direction = "fr_to_en"
-        elif source_lang == "en" and target_lang == "fr":
-            direction = "en_to_fr"
-        else:
-            return f"Traduction non prise en charge: {source_lang} vers {target_lang}"
+        # Utiliser les langues spécifiées ou celles de la configuration
+        src_lang = source_language or self.config["source_language"]
+        tgt_lang = target_language or self.config["target_language"]
         
-        # Construire le prompt
-        prompt = self.prompts[direction].format(text=text)
+        # Génération d'un ID unique pour cette traduction
+        translation_id = str(uuid.uuid4())[:8]
+        timestamp = time.strftime("%Y%m%d-%H%M%S")
+        
+        # Construire le prompt pour le modèle
+        system_prompt = f"You are a professional and accurate translator. Translate the following text "
+        system_prompt += f"from {src_lang} to {tgt_lang}. Maintain the original formatting and structure as much as possible."
+        
+        # Construire le message utilisateur avec le texte à traduire
+        user_prompt = prompt
+        
+        # Créer l'API Ollama pour l'appel direct
+        api = OllamaAPI(base_url=self.endpoint)
+        
+        # Journaliser le prompt complet
+        full_prompt = f"System: {system_prompt}\n\nUser: {user_prompt}"
+        print(f"Envoi du prompt de traduction au modèle {self.model_name}:\n{system_prompt}")
         
         try:
-            # Construire la payload pour l'API Ollama
-            payload = {
-                "model": self.model_name,
-                "prompt": prompt,
-                "options": {
-                    "temperature": self.config.get("temperature", 0.1),  # Basse pour traduction précise
+            # Pour les modèles qui supportent le format de chat
+            if any(name in self.model_name.lower() for name in ["llama", "mistral", "deepseek", "qwen"]):
+                # Formater en tant que messages de chat
+                messages = [
+                    {"role": "system", "content": system_prompt},
+                    {"role": "user", "content": user_prompt}
+                ]
+                
+                response = api.chat(
+                    model=self.model_name,
+                    messages=messages,
+                    options={
+                        "temperature": self.config.get("temperature", 0.1),  # Température plus basse pour la traduction
                         "top_p": self.config.get("top_p", 0.95),
                         "top_k": self.config.get("top_k", 40),
                         "num_predict": self.config.get("max_tokens", 2048)
                     }
-            }
+                )
                 
-            # Dans une implémentation réelle, envoyer la requête à l'API Ollama
-            # response = requests.post(f"{self.endpoint}/api/generate", json=payload)
-            # json_response = response.json()
-            # return json_response.get("response", "")
-            
-            # Pour cette démonstration, retourner une traduction simulée
-            if direction == "fr_to_en":
-                if "schéma" in text.lower():
-                    return text.replace("schéma", "diagram").replace("Schéma", "Diagram")
-                elif "tableau" in text.lower():
-                    return text.replace("tableau", "table").replace("Tableau", "Table")
-                elif "formule" in text.lower():
-                    return text.replace("formule", "formula").replace("Formule", "Formula")
+                if "message" in response and "content" in response["message"]:
+                    result = response["message"]["content"]
                 else:
-                    # Exemple très simplifié de "traduction"
-                    translations = {
-                        "Le": "The", "la": "the", "les": "the", "des": "the",
-                        "et": "and", "ou": "or", "pour": "for", "avec": "with",
-                        "est": "is", "sont": "are", "contient": "contains",
-                        "montre": "shows", "représente": "represents",
-                        "plusieurs": "several", "important": "important",
-                        "information": "information", "données": "data",
-                        "processus": "process", "système": "system",
-                        "analyse": "analysis", "résultat": "result"
-                    }
-                    
-                    # Remplacement simple mot à mot
-                    result = text
-                    for fr, en in translations.items():
-                        result = result.replace(f" {fr} ", f" {en} ")
-                    
-                    return result
-            else:  # en_to_fr
-                if "diagram" in text.lower():
-                    return text.replace("diagram", "schéma").replace("Diagram", "Schéma")
-                elif "table" in text.lower():
-                    return text.replace("table", "tableau").replace("Table", "Tableau")
-                elif "formula" in text.lower():
-                    return text.replace("formula", "formule").replace("Formula", "Formule")
+                    result = response.get("response", "Erreur: Format de réponse inattendu")
             else:
-                    # Exemple très simplifié de "traduction"
-                    translations = {
-                        "The": "Le", "the": "le", "and": "et", "or": "ou",
-                        "for": "pour", "with": "avec", "is": "est",
-                        "are": "sont", "contains": "contient", "shows": "montre",
-                        "represents": "représente", "several": "plusieurs",
-                        "important": "important", "information": "information",
-                        "data": "données", "process": "processus",
-                        "system": "système", "analysis": "analyse",
-                        "result": "résultat"
+                # Format de génération standard pour les autres modèles
+                prompt_text = f"{system_prompt}\n\n{user_prompt}"
+                
+                response = api.generate(
+                    model=self.model_name,
+                    prompt=prompt_text,
+                    options={
+                        "temperature": self.config.get("temperature", 0.1),
+                        "top_p": self.config.get("top_p", 0.95),
+                        "top_k": self.config.get("top_k", 40),
+                        "num_predict": self.config.get("max_tokens", 2048)
                     }
+                )
                 
-                    # Remplacement simple mot à mot
-                    result = text
-                    for en, fr in translations.items():
-                        result = result.replace(f" {en} ", f" {fr} ")
+                result = response.get("response", "Erreur: Pas de réponse")
             
+            # Enregistrer la traduction dans un fichier
+            translation_path = os.path.join(self.log_dir, f"{timestamp}_{translation_id}.txt")
+            with open(translation_path, "w", encoding="utf-8") as f:
+                f.write(f"Source language: {src_lang}\n")
+                f.write(f"Target language: {tgt_lang}\n")
+                f.write(f"Model: {self.model_name}\n\n")
+                f.write(f"Original text:\n{prompt}\n\n")
+                f.write(f"Translation:\n{result}")
+            
+            print(f"Traduction enregistrée dans: {translation_path}")
             return result
             
         except Exception as e:
-            return f"Erreur lors de la traduction: {str(e)}" 
+            error_msg = f"Erreur lors de la traduction: {str(e)}"
+            print(error_msg)
+            
+            # Enregistrer l'erreur
+            error_path = os.path.join(self.log_dir, f"{timestamp}_{translation_id}_error.txt")
+            with open(error_path, "w", encoding="utf-8") as f:
+                f.write(f"Source language: {src_lang}\n")
+                f.write(f"Target language: {tgt_lang}\n")
+                f.write(f"Model: {self.model_name}\n\n")
+                f.write(f"Original text:\n{prompt}\n\n")
+                f.write(f"Error:\n{str(e)}")
+                
+            return error_msg 

agents/vision.py

@@ -2,145 +2,192 @@
 # -*- coding: utf-8 -*-
 
 """
-Agent LLM pour l'analyse visuelle d'images
+Agent pour l'analyse d'images et de schémas
 """
 
-import base64
-import json
-import requests
-from typing import List, Optional, Union
+import os
+import time
+import uuid
+from PIL import Image
+import io
+from typing import List, Optional, Dict, Any
 
 from .base import LLMBaseAgent
+from utils.api_ollama import OllamaAPI
 
 class VisionAgent(LLMBaseAgent):
     """
-    Agent LLM spécialisé dans l'analyse d'images
+    Agent pour l'analyse d'images avec des modèles multimodaux
     """
     
-    def __init__(self, model_name: str = "llama3.2-vision:90b", **config):
+    def __init__(self, model_name: str, endpoint: str = "http://217.182.105.173:11434", **config):
         """
         Initialise l'agent de vision
         
         Args:
-            model_name (str): Nom du modèle de vision (défaut: llama3.2-vision:90b)
-            **config: Paramètres de configuration supplémentaires
+            model_name (str): Nom du modèle à utiliser
+            endpoint (str): URL de l'API Ollama
+            **config: Configuration supplémentaire
         """
-        super().__init__(model_name, **config)
+        super().__init__(model_name, endpoint, **config)
         
-        # Définir les prompts par défaut selon la langue
-        self.prompts = {
-            "fr": {
-                "schéma": "Décris en détail ce schéma. Quelle information principale est-elle représentée ? "
-                          "Comment les éléments sont-ils organisés ? Quelle est la signification des différentes parties ?",
-                
-                "tableau": "Analyse ce tableau. Quel type de données contient-il ? "
-                           "Quelles sont les informations importantes ? Résume son contenu et sa structure.",
-                
-                "formule": "Analyse cette formule ou équation mathématique. "
-                           "Que représente-t-elle ? Quelle est sa signification et son application ?",
-                
-                "texte": "Lis et résume ce texte. Quel est le contenu principal ? "
-                         "Quels sont les points importants à retenir ?",
-                
-                "autre": "Décris en détail ce que tu vois sur cette image. "
-                         "Quel est le contenu principal ? Quelle information est présentée ?"
-            },
-            "en": {
-                "schéma": "Describe this diagram in detail. What main information is represented? "
-                          "How are the elements organized? What is the meaning of the different parts?",
-                
-                "tableau": "Analyze this table. What kind of data does it contain? "
-                           "What is the important information? Summarize its content and structure.",
-                
-                "formule": "Analyze this mathematical formula or equation. "
-                           "What does it represent? What is its meaning and application?",
-                
-                "texte": "Read and summarize this text. What is the main content? "
-                         "What are the important points to remember?",
-                
-                "autre": "Describe in detail what you see in this image. "
-                         "What is the main content? What information is presented?"
-            }
+        # Configuration par défaut
+        default_config = {
+            "save_images": True  # Enregistrer les images par défaut
         }
         
+        # Mettre à jour la configuration avec les valeurs par défaut si non spécifiées
+        for key, value in default_config.items():
+            if key not in self.config:
+                self.config[key] = value
+        
+        # Création du répertoire pour sauvegarder les images analysées
+        self.image_dir = os.path.join("data", "images")
+        os.makedirs(self.image_dir, exist_ok=True)
+    
     def generate(self, prompt: Optional[str] = "", images: Optional[List[bytes]] = None, 
                   selection_type: str = "autre", context: Optional[str] = "") -> str:
         """
         Génère une description ou une analyse d'une image
         
         Args:
-            prompt (str, optional): Prompt personnalisé (si vide, utilise le prompt par défaut)
-            images (List[bytes], optional): Liste d'images en bytes
-            selection_type (str): Type de contenu ("schéma", "tableau", "formule", "texte", "autre")
-            context (str, optional): Contexte textuel associé à la sélection
+            prompt (str, optional): Prompt supplémentaire (non utilisé)
+            images (List[bytes], optional): Liste d'images à analyser
+            selection_type (str): Type de la sélection (schéma, tableau, formule...)
+            context (str, optional): Contexte textuel
             
         Returns:
-            str: La réponse générée par le modèle
+            str: Description générée par le modèle
         """
-        if not images:
-            return "Aucune image fournie pour l'analyse."
+        if not images or len(images) == 0:
+            return "Erreur: Aucune image fournie pour l'analyse"
         
-        # Utiliser le prompt par défaut si aucun n'est fourni
-        if not prompt:
-            lang = self.config.get("language", "fr")
-            prompts_dict = self.prompts.get(lang, self.prompts["fr"])
-            prompt = prompts_dict.get(selection_type, prompts_dict["autre"])
+        image_data = images[0]
         
-            # Ajouter le contexte si disponible
-            if context:
-                prompt = f"Contexte: {context}\n\n{prompt}"
+        # Sauvegarder l'image pour référence future seulement si l'option est activée
+        image_id = str(uuid.uuid4())[:8]
+        timestamp = time.strftime("%Y%m%d-%H%M%S")
+        image_filename = f"{timestamp}_{image_id}.png"
+        image_path = os.path.join(self.image_dir, image_filename)
+        
+        if self.config.get("save_images", True):
+            try:
+                # Sauvegarder l'image
+                img = Image.open(io.BytesIO(image_data))
+                img.save(image_path)
+                print(f"Image sauvegardée: {image_path}")
+            except Exception as e:
+                print(f"Erreur lors de la sauvegarde de l'image: {str(e)}")
+        
+        # Construction du prompt en anglais pour le modèle
+        system_prompt = "Analyze the following image"
+        
+        # Mapper les types de sélection en français vers l'anglais
+        content_type_mapping = {
+            "schéma": "diagram",
+            "tableau": "table",
+            "formule": "formula",
+            "graphique": "chart",
+            "autre": "content"
+        }
+        
+        # Obtenir le type en anglais ou utiliser le type original
+        content_type_en = content_type_mapping.get(selection_type.lower(), selection_type)
+        system_prompt += f" which contains {content_type_en}"
+        
+        # Ajout d'instructions spécifiques selon le type de contenu
+        if content_type_en == "diagram":
+            system_prompt += ". Please describe in detail what this diagram shows, including all components, connections, and what it represents."
+        elif content_type_en == "table":
+            system_prompt += ". Please extract and format the table content, describing its structure, headers, and data. If possible, recreate the table structure."
+        elif content_type_en == "formula" or content_type_en == "equation":
+            system_prompt += ". Please transcribe this mathematical formula/equation and explain what it represents and its components."
+        elif content_type_en == "chart" or content_type_en == "graph":
+            system_prompt += ". Please describe this chart/graph in detail, including the axes, data points, trends, and what information it conveys."
+        else:
+            system_prompt += ". Please provide a detailed description of what you see in this image."
+        
+        # Ajouter des instructions générales
+        system_prompt += "\n\nPlease be detailed and precise in your analysis."
+        
+        # Préparer le prompt avec le contexte
+        user_prompt = ""
+        if context and context.strip():
+            # Le contexte est déjà en français, pas besoin de le traduire
+            # mais préciser explicitement que c'est en français pour le modèle
+            user_prompt = f"Here is additional context that may help with your analysis (may be in French):\n{context}"
+        
+        # Créer l'API Ollama pour l'appel direct
+        api = OllamaAPI(base_url=self.endpoint)
+        
+        # Journaliser le prompt complet
+        full_prompt = f"System: {system_prompt}\n\nUser: {user_prompt}"
+        print(f"Envoi du prompt au modèle {self.model_name}:\n{full_prompt}")
         
         try:
-            # Pour chaque image, encoder en base64
-            base64_images = []
-            for image in images:
-                if isinstance(image, bytes):
-                    base64_image = base64.b64encode(image).decode("utf-8")
-                    base64_images.append(base64_image)
+            # Pour les modèles qui supportent le format de chat
+            if "llama" in self.model_name.lower() or "llava" in self.model_name.lower():
+                # Formater en tant que messages de chat
+                messages = [
+                    {"role": "system", "content": system_prompt}
+                ]
                 
-            # Construire la payload pour l'API Ollama
-            payload = {
-                "model": self.model_name,
-                "prompt": prompt,
-                "images": base64_images,
-                "options": {
+                if user_prompt:
+                    messages.append({"role": "user", "content": user_prompt})
+                
+                response = api.chat(
+                    model=self.model_name,
+                    messages=messages,
+                    images=[image_data],
+                    options={
                         "temperature": self.config.get("temperature", 0.2),
                         "top_p": self.config.get("top_p", 0.95),
                         "top_k": self.config.get("top_k", 40),
                         "num_predict": self.config.get("max_tokens", 1024)
                     }
-            }
-            
-            # Dans une implémentation réelle, envoyer la requête à l'API Ollama
-            # response = requests.post(f"{self.endpoint}/api/generate", json=payload)
-            # json_response = response.json()
-            # return json_response.get("response", "")
-            
-            # Pour cette démonstration, retourner une réponse simulée
-            if selection_type == "schéma":
-                return "Le schéma présenté illustre un processus structuré avec plusieurs étapes interconnectées. " \
-                       "On peut observer une organisation hiérarchique des éléments, avec des flèches indiquant " \
-                       "le flux d'information ou la séquence d'opérations. Les différentes composantes sont " \
-                       "clairement délimitées et semblent représenter un workflow ou un système de classification."
-            
-            elif selection_type == "tableau":
-                return "Ce tableau contient plusieurs colonnes et rangées de données structurées. " \
-                       "Il présente une organisation systématique d'informations, probablement des " \
-                       "valeurs numériques ou des catégories. Les en-têtes indiquent le type de données " \
-                       "dans chaque colonne, et l'ensemble forme une matrice cohérente d'informations liées."
-            
-            elif selection_type == "formule":
-                return "Cette formule mathématique représente une relation complexe entre plusieurs variables. " \
-                       "Elle utilise divers opérateurs et symboles mathématiques pour exprimer un concept ou " \
-                       "une règle. La structure suggère qu'il s'agit d'une équation importante dans son domaine, " \
-                       "possiblement liée à un phénomène physique ou à un modèle théorique."
+                )
                 
+                if "message" in response and "content" in response["message"]:
+                    result = response["message"]["content"]
                 else:
-                return "L'image montre un contenu visuel structuré qui présente des informations importantes " \
-                       "dans le contexte du document. Les éléments visuels sont organisés de manière à faciliter " \
-                       "la compréhension d'un concept ou d'un processus spécifique. La qualité et la disposition " \
-                       "des éléments suggèrent qu'il s'agit d'une représentation professionnelle destinée à " \
-                       "communiquer efficacement l'information."
+                    result = response.get("response", "Erreur: Format de réponse inattendu")
+            else:
+                # Format de génération standard pour les autres modèles
+                prompt_text = system_prompt
+                if user_prompt:
+                    prompt_text += f"\n\n{user_prompt}"
+                
+                response = api.generate(
+                    model=self.model_name,
+                    prompt=prompt_text,
+                    images=[image_data],
+                    options={
+                        "temperature": self.config.get("temperature", 0.2),
+                        "top_p": self.config.get("top_p", 0.95),
+                        "top_k": self.config.get("top_k", 40),
+                        "num_predict": self.config.get("max_tokens", 1024)
+                    }
+                )
+                
+                result = response.get("response", "Erreur: Pas de réponse")
+            
+            # Enregistrer la réponse dans un fichier si l'option d'enregistrement est activée
+            if self.config.get("save_images", True):
+                response_path = os.path.join(self.image_dir, f"{timestamp}_{image_id}_response.txt")
+                with open(response_path, "w", encoding="utf-8") as f:
+                    f.write(f"Prompt:\n{full_prompt}\n\nResponse:\n{result}")
+                
+                print(f"Réponse enregistrée dans: {response_path}")
+            return result
             
         except Exception as e:
-            return f"Erreur lors de l'analyse de l'image: {str(e)}" 
+            error_msg = f"Erreur lors de l'analyse de l'image: {str(e)}"
+            print(error_msg)
+            
+            # Enregistrer l'erreur si l'option d'enregistrement est activée
+            if self.config.get("save_images", True):
+                error_path = os.path.join(self.image_dir, f"{timestamp}_{image_id}_error.txt")
+                with open(error_path, "w", encoding="utf-8") as f:
+                    f.write(f"Prompt:\n{full_prompt}\n\nError:\n{str(e)}")
+                
+            return error_msg 

check_prerequisites.ps1

@@ -0,0 +1,154 @@
+# Prerequisites check script for Ragflow PDF Preprocessing
+Write-Host "======================================================" -ForegroundColor Cyan
+Write-Host "   Ragflow PDF Preprocessing Prerequisites Check" -ForegroundColor Cyan
+Write-Host "======================================================" -ForegroundColor Cyan
+Write-Host ""
+
+$allOk = $true
+
+# Check Python and version
+try {
+    $pythonVersionOutput = (python --version) 2>&1
+    $pythonVersion = $pythonVersionOutput -replace "Python "
+    
+    if ([version]$pythonVersion -ge [version]"3.8") {
+        Write-Host "✅ Python $pythonVersion is installed (>= 3.8 required)" -ForegroundColor Green
+    } else {
+        Write-Host "❌ Python $pythonVersion is installed, but version 3.8 or higher is required" -ForegroundColor Red
+        $allOk = $false
+    }
+} catch {
+    Write-Host "❌ Python is not installed or not in PATH" -ForegroundColor Red
+    Write-Host "   Required installation: https://www.python.org/downloads/" -ForegroundColor Yellow
+    Write-Host "   Make sure to check 'Add Python to PATH' during installation" -ForegroundColor Yellow
+    $allOk = $false
+}
+
+# Check Tesseract OCR
+$tesseractPaths = @(
+    "C:\Program Files\Tesseract-OCR\tesseract.exe",
+    "C:\Program Files (x86)\Tesseract-OCR\tesseract.exe",
+    "C:\Tesseract-OCR\tesseract.exe"
+)
+
+$tesseractInstalled = $false
+$tesseractPath = ""
+
+foreach ($path in $tesseractPaths) {
+    if (Test-Path $path) {
+        $tesseractInstalled = $true
+        $tesseractPath = $path
+        break
+    }
+}
+
+if ($tesseractInstalled) {
+    Write-Host "✅ Tesseract OCR is installed at: $tesseractPath" -ForegroundColor Green
+    
+    # Check language packs
+    $tesseractDir = Split-Path -Parent $tesseractPath
+    $langPath = Join-Path $tesseractDir "tessdata"
+    $fraLangPath = Join-Path $langPath "fra.traineddata"
+    $engLangPath = Join-Path $langPath "eng.traineddata"
+    
+    if (Test-Path $fraLangPath) {
+        Write-Host "  ✅ French language pack (fra) installed" -ForegroundColor Green
+    } else {
+        Write-Host "  ❌ French language pack (fra) missing" -ForegroundColor Yellow
+        Write-Host "     Reinstall Tesseract selecting the language packs" -ForegroundColor Yellow
+        $allOk = $false
+    }
+    
+    if (Test-Path $engLangPath) {
+        Write-Host "  ✅ English language pack (eng) installed" -ForegroundColor Green
+    } else {
+        Write-Host "  ❌ English language pack (eng) missing" -ForegroundColor Yellow
+        Write-Host "     Reinstall Tesseract selecting the language packs" -ForegroundColor Yellow
+        $allOk = $false
+    }
+} else {
+    Write-Host "❌ Tesseract OCR is not installed" -ForegroundColor Red
+    Write-Host "   Required installation: https://github.com/UB-Mannheim/tesseract/wiki" -ForegroundColor Yellow
+    Write-Host "   Make sure to select French and English language packs" -ForegroundColor Yellow
+    $allOk = $false
+}
+
+# Check Ollama server
+try {
+    $response = Invoke-WebRequest -Uri "http://217.182.105.173:11434/api/version" -UseBasicParsing -ErrorAction SilentlyContinue
+    if ($response.StatusCode -eq 200) {
+        Write-Host "✅ Ollama server is accessible at 217.182.105.173:11434" -ForegroundColor Green
+        
+        # Check available models
+        $modelsResponse = Invoke-WebRequest -Uri "http://217.182.105.173:11434/api/tags" -UseBasicParsing -ErrorAction SilentlyContinue
+        $models = ($modelsResponse.Content | ConvertFrom-Json).models
+        $modelNames = $models | ForEach-Object { $_.name }
+        
+        # Required models based on the server's available models
+        $requiredModels = @(
+            "mistral:latest",
+            "llava:34b-v1.6-fp16", 
+            "llama3.2-vision:90b-instruct-q8_0"
+        )
+        
+        foreach ($model in $requiredModels) {
+            if ($modelNames -contains $model) {
+                Write-Host "  ✅ Model $model is available" -ForegroundColor Green
+            } else {
+                Write-Host "  ❌ Model $model not found on server" -ForegroundColor Yellow
+                Write-Host "     This model is required for some features" -ForegroundColor Yellow
+                $allOk = $false
+            }
+        }
+    }
+} catch {
+    Write-Host "❌ Cannot connect to Ollama server at 217.182.105.173:11434" -ForegroundColor Red
+    Write-Host "   Make sure you have network connectivity to the Ollama server" -ForegroundColor Yellow
+    Write-Host "   This is required for the LLM features of the application" -ForegroundColor Yellow
+    $allOk = $false
+}
+
+# Check Python libraries
+Write-Host ""
+Write-Host "Checking required Python libraries:" -ForegroundColor Cyan
+
+$requiredLibraries = @(
+    "PyQt6",
+    "PyMuPDF",
+    "numpy",
+    "pytesseract",
+    "Pillow",
+    "opencv-python",
+    "requests"
+)
+
+foreach ($lib in $requiredLibraries) {
+    try {
+        $output = python -c "import $($lib.ToLower().Replace('-', '_')); print('OK')" 2>&1
+        if ($output -eq "OK") {
+            Write-Host "  ✅ $lib is installed" -ForegroundColor Green
+        } else {
+            Write-Host "  ❌ $lib is not installed correctly" -ForegroundColor Red
+            Write-Host "     Recommended installation: pip install $lib" -ForegroundColor Yellow
+            $allOk = $false
+        }
+    } catch {
+        Write-Host "  ❌ $lib is not installed" -ForegroundColor Red
+        Write-Host "     Recommended installation: pip install $lib" -ForegroundColor Yellow
+        $allOk = $false
+    }
+}
+
+# Summary
+Write-Host ""
+Write-Host "======================================================" -ForegroundColor Cyan
+if ($allOk) {
+    Write-Host "✅ All prerequisites are satisfied!" -ForegroundColor Green
+    Write-Host "   You can proceed with the installation and use of Ragflow." -ForegroundColor Green
+} else {
+    Write-Host "⚠️ Some prerequisites are not satisfied." -ForegroundColor Yellow
+    Write-Host "   Please install the missing components before using Ragflow." -ForegroundColor Yellow
+}
+Write-Host "======================================================" -ForegroundColor Cyan
+
+Read-Host "Press ENTER to exit" 

config/__init__.py

@@ -0,0 +1,8 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+
+"""
+Package de configuration de l'application
+"""
+
+# Ce fichier permet d'importer le package config 

config/agent_config.py

@@ -0,0 +1,54 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+
+"""
+Configuration des agents dans l'application
+"""
+
+# Configuration des agents activés/désactivés
+ACTIVE_AGENTS = {
+    "ocr": True,         # Agent de reconnaissance optique de caractères
+    "vision": True,      # Agent d'analyse d'images
+    "translation": True, # Agent de traduction
+    "summary": False,    # Agent de résumé (désactivé)
+    "rewriter": True     # Agent de reformulation
+}
+
+# Configuration des modèles par défaut pour chaque agent
+DEFAULT_MODELS = {
+    "ocr": None,  # OCR utilise Tesseract, pas de modèle LLM
+    "vision": "llava:34b-v1.6-fp16",  # Modèle par défaut pour l'analyse d'images
+    "translation": "mistral:latest", # Modèle par défaut pour la traduction
+    "summary": "mistral:latest",     # Modèle par défaut pour le résumé
+    "rewriter": "mistral:latest"     # Modèle par défaut pour la reformulation
+}
+
+# Configuration de l'endpoint Ollama par défaut
+DEFAULT_ENDPOINT = "http://217.182.105.173:11434"
+
+# Configuration de journalisation détaillée
+VERBOSE_LOGGING = True  # Enregistre tous les détails des entrées/sorties
+
+def is_agent_enabled(agent_name: str) -> bool:
+    """
+    Vérifie si un agent est activé dans la configuration
+    
+    Args:
+        agent_name (str): Nom de l'agent à vérifier
+        
+    Returns:
+        bool: True si l'agent est activé, False sinon
+    """
+    return ACTIVE_AGENTS.get(agent_name, False)
+
+def get_default_model(agent_name: str) -> str:
+    """
+    Renvoie le modèle par défaut pour un agent donné
+    
+    Args:
+        agent_name (str): Nom de l'agent
+        
+    Returns:
+        str: Nom du modèle par défaut
+    """
+    return DEFAULT_MODELS.get(agent_name, "mistral") 

config/llm_profiles.json

@@ -1,7 +1,7 @@
 {
   "léger": {
     "vision": {
-      "model": "llava:34b",
+      "model": "llava:34b-v1.6-fp16",
       "language": "en",
       "temperature": 0.2,
       "top_p": 0.95,
@@ -9,7 +9,7 @@
       "max_tokens": 1024
     },
     "translation": {
-      "model": "mistral",
+      "model": "mistral:latest",
       "language": "fr",
       "temperature": 0.1,
       "top_p": 0.95,
@@ -17,7 +17,7 @@
       "max_tokens": 1024
     },
     "summary": {
-      "model": "mistral",
+      "model": "mistral:latest",
       "language": "fr",
       "temperature": 0.2,
       "top_p": 0.95,
@@ -25,7 +25,7 @@
       "max_tokens": 1024
     },
     "rewriter": {
-      "model": "mistral",
+      "model": "mistral:latest",
       "language": "fr",
       "temperature": 0.3,
       "top_p": 0.95,
@@ -35,7 +35,7 @@
   },
   "moyen": {
     "vision": {
-      "model": "llava",
+      "model": "llava:34b-v1.6-fp16",
       "language": "en",
       "temperature": 0.2,
       "top_p": 0.95,
@@ -43,7 +43,7 @@
       "max_tokens": 1024
     },
     "translation": {
-      "model": "qwen2.5",
+      "model": "qwen2.5:72b-instruct-q8_0",
       "language": "fr",
       "temperature": 0.1,
       "top_p": 0.95,
@@ -51,7 +51,7 @@
       "max_tokens": 1024
     },
     "summary": {
-      "model": "deepseek-r1",
+      "model": "deepseek-r1:70b-llama-distill-q8_0",
       "language": "fr",
       "temperature": 0.2,
       "top_p": 0.95,
@@ -59,7 +59,7 @@
       "max_tokens": 1024
     },
     "rewriter": {
-      "model": "mistral",
+      "model": "mistral:latest",
       "language": "fr",
       "temperature": 0.3,
       "top_p": 0.95,
@@ -69,7 +69,7 @@
   },
   "avancé": {
     "vision": {
-      "model": "llama3.2-vision",
+      "model": "llama3.2-vision:90b-instruct-q8_0",
       "language": "en",
       "temperature": 0.2,
       "top_p": 0.95,
@@ -77,7 +77,7 @@
       "max_tokens": 2048
     },
     "translation": {
-      "model": "deepseek",
+      "model": "deepseek-r1:70b-llama-distill-q8_0",
       "language": "fr",
       "temperature": 0.1,
       "top_p": 0.95,
@@ -85,7 +85,7 @@
       "max_tokens": 2048
     },
     "summary": {
-      "model": "deepseek-r1",
+      "model": "deepseek-r1:70b-llama-distill-q8_0",
       "language": "fr",
       "temperature": 0.2,
       "top_p": 0.95,
@@ -93,7 +93,7 @@
       "max_tokens": 2048
     },
     "rewriter": {
-      "model": "deepseek",
+      "model": "deepseek-r1:70b-llama-distill-q8_0",
       "language": "fr",
       "temperature": 0.3,
       "top_p": 0.95,

data/images/.gitkeep

@@ -0,0 +1 @@
+# Ce fichier permet de conserver le répertoire vide dans Git

data/images/20250327-175335_d74b588a_response.txt

@@ -0,0 +1,10 @@
+Prompt:
+System: Analyze the following image which contains diagram. Please describe in detail what this diagram shows, including all components, connections, and what it represents.
+
+Please be detailed and precise in your analysis.
+
+User: Here is additional context that may help with your analysis (may be in French):
+L = plus grande longueur - D = plus grand diamètre (permettant de passer une maille de tamis £ D). - 1 = dimension perpendiculaire au plan LD et inférieure à D Figure 1 : définition de la Lmax
+
+Response:
+Error: model 'llava:34b' not found, try pulling it first

data/ocr_logs/.gitkeep

@@ -0,0 +1 @@
+# Ce fichier permet de conserver le répertoire vide dans Git

data/outputs/.gitkeep

@@ -0,0 +1 @@
+# Ce fichier permet de conserver le répertoire vide dans Git

data/outputs/analyse_document.md

@@ -0,0 +1,9 @@
+# Analyse de document
+
+## Schéma - Page 8
+
+**Contexte** :
+L = plus grande longueur - D = plus grand diamètre (permettant de passer une maille de tamis £ D). - 1 = dimension perpendiculaire au plan LD et inférieure à D Figure 1 : définition de la Lmax
+
+---
+

data/rewrites/.gitkeep

@@ -0,0 +1 @@
+# Ce fichier permet de conserver le répertoire vide dans Git

data/summaries/.gitkeep

@@ -0,0 +1 @@
+# Ce fichier permet de conserver le répertoire vide dans Git

data/temp/.gitkeep

@@ -0,0 +1 @@
+# Ce fichier permet de conserver le répertoire vide dans Git

data/templates/.gitkeep

@@ -0,0 +1 @@
+# Ce fichier permet de conserver le répertoire vide dans Git

data/translations/.gitkeep

@@ -0,0 +1 @@
+# Ce fichier permet de conserver le répertoire vide dans Git

data/translations/20250327-175336_b6e18fc4.txt

@@ -0,0 +1,14 @@
+Source language: en
+Target language: fr
+Model: mistral
+
+Original text:
+Voici le contexte de l'image: L = plus grande longueur - D = plus grand diamètre (permettant de passer une maille de tamis £ D). - 1 = dimension perpendiculaire au plan LD et inférieure à D Figure 1 : définition de la Lmax
+
+Translation:
+ Voici le contexte de l'image :
+
+L = longueur maximale ; D = diamètre maximal (permettant de passer une maille de tamis £ D) ;
+1 = dimension perpendiculaire au plan LD et inférieure à D
+
+Figure 1 : définition de la Lmax

data/uploads/.gitkeep

@@ -0,0 +1 @@
+# Ce fichier permet de conserver le répertoire vide dans Git

data/workflows/.gitkeep

@@ -0,0 +1 @@
+# Ce fichier permet de conserver le répertoire vide dans Git

data/workflows/20250327-175335_session_1743094415_workflow.md

@@ -0,0 +1,62 @@
+# Flux de traitement - Session session_1743094415
+
+Date et heure: 27/03/2025 17:53:35
+
+## Étapes du traitement
+
+### Étape: vision (17:53:35)
+
+#### Métadonnées
+
+- **model**: llava:34b
+
+#### Entrées
+
+- **content_type**:
+```
+schéma
+```
+- **context**:
+```
+L = plus grande longueur - D = plus grand diamètre (permettant de passer une maille de tamis £ D). - 1 = dimension perpendiculaire au plan LD et inférieure à D Figure 1 : définition de la Lmax
+```
+#### Sorties
+
+```
+Error: model 'llava:34b' not found, try pulling it first
+```
+
+---
+
+### Étape: translation (17:53:39)
+
+#### Métadonnées
+
+- **model**: mistral
+
+#### Entrées
+
+- **text**:
+```
+Voici le contexte de l'image: L = plus grande longueur - D = plus grand diamètre (permettant de passer une maille de tamis £ D). - 1 = dimension perpendiculaire au plan LD et inférieure à D Figure 1 : définition de la Lmax
+```
+#### Sorties
+
+```
+ Voici le contexte de l'image :
+
+L = longueur maximale ; D = diamètre maximal (permettant de passer une maille de tamis £ D) ;
+1 = dimension perpendiculaire au plan LD et inférieure à D
+
+Figure 1 : définition de la Lmax
+```
+
+---
+
+## Résumé du traitement
+
+Traitement terminé avec succès.
+Modèles utilisés: vision=llava:34b, translation=mistral
+Agents exécutés: vision, translation
+
+*Fin du flux de traitement - 27/03/2025 17:53:40*

debug_setup.bat

@@ -0,0 +1,53 @@
+@echo off
+:: Debug and setup script for RagFlow Preprocess
+echo ===================================================
+echo  RagFlow Preprocess Setup and Debug
+echo ===================================================
+echo.
+
+echo Installing Python dependencies...
+pip install requests PyQt6 PyPDF2 pytesseract Pillow PyMuPDF
+echo.
+
+echo Checking Tesseract OCR installation...
+:: Check if Tesseract is installed
+where tesseract >nul 2>nul
+if %ERRORLEVEL% neq 0 (
+    echo WARNING: Tesseract is not installed or not in PATH.
+    echo Please download and install Tesseract OCR from:
+    echo https://github.com/UB-Mannheim/tesseract/wiki
+    echo.
+    echo IMPORTANT: During installation, check the option to add Tesseract to PATH.
+    echo            And make sure to install French and English languages.
+    echo.
+    echo Press any key to continue...
+    pause >nul
+)
+
+echo Testing communication with Ollama server...
+python test_components.py
+echo.
+
+echo Updating configurations...
+echo Ollama Server: http://217.182.105.173:11434
+
+echo.
+echo ===================================================
+echo  FINAL INSTRUCTIONS
+echo ===================================================
+echo.
+echo 1. If Tesseract OCR is not installed yet:
+echo    - Download it from: https://github.com/UB-Mannheim/tesseract/wiki
+echo    - Install it with the "Add to PATH" option checked
+echo    - Make sure to install French (fra) and English (eng) languages
+echo.
+echo 2. To install French languages for Tesseract:
+echo    - Launch PowerShell as administrator
+echo    - Run the script: .\install_languages.ps1
+echo.
+echo 3. To run the application:
+echo    - Use: python main.py
+echo.
+
+echo Press any key to exit...
+pause >nul 

export_cursor_chats_to_md.py

@@ -1,49 +0,0 @@
-import os
-import json
-from datetime import datetime
-
-# === Config ===
-CURSOR_CHAT_DIR = os.path.expanduser("~/.cursor/chat/")
-OUTPUT_FILE = "cursor_history.md"
-
-# === Initialisation du contenu ===
-md_output = ""
-
-# === Chargement des discussions Cursor ===
-for filename in sorted(os.listdir(CURSOR_CHAT_DIR)):
-    if not filename.endswith(".json"):
-        continue
-
-    filepath = os.path.join(CURSOR_CHAT_DIR, filename)
-
-    with open(filepath, "r", encoding="utf-8") as f:
-        try:
-            chat_data = json.load(f)
-        except json.JSONDecodeError:
-            continue  # Fichier corrompu ou non lisible
-
-    created_at_raw = chat_data.get("createdAt", "")
-    try:
-        created_at = datetime.fromisoformat(created_at_raw.replace("Z", ""))
-    except ValueError:
-        created_at = datetime.now()
-
-    formatted_time = created_at.strftime("%Y-%m-%d %H:%M:%S")
-    md_output += f"\n---\n\n## Session du {formatted_time}\n\n"
-
-    for msg in chat_data.get("messages", []):
-        role = msg.get("role", "")
-        content = msg.get("content", "").strip()
-        if not content:
-            continue
-
-        if role == "user":
-            md_output += f"** Utilisateur :**\n{content}\n\n"
-        elif role == "assistant":
-            md_output += f"** Assistant :**\n{content}\n\n"
-
-# === Écriture / ajout dans le fichier final ===
-with open(OUTPUT_FILE, "a", encoding="utf-8") as output_file:
-    output_file.write(md_output)
-
-print(f" Export terminé ! Discussions ajoutées à : {OUTPUT_FILE}")

install_languages.ps1

@@ -0,0 +1,86 @@
+#!/usr/bin/env pwsh
+
+# Script to install additional languages for Tesseract OCR
+
+# Check if the script is running as administrator
+if (-NOT ([Security.Principal.WindowsPrincipal][Security.Principal.WindowsIdentity]::GetCurrent()).IsInRole([Security.Principal.WindowsBuiltInRole] "Administrator")) {
+    Write-Warning "Please run this script as administrator!"
+    Write-Host "The script will close in 5 seconds..."
+    Start-Sleep -Seconds 5
+    exit
+}
+
+# Function to find Tesseract installation path
+function Find-TesseractPath {
+    $possiblePaths = @(
+        "C:\Program Files\Tesseract-OCR",
+        "C:\Program Files (x86)\Tesseract-OCR",
+        "C:\Tesseract-OCR"
+    )
+    
+    foreach ($path in $possiblePaths) {
+        if (Test-Path "$path\tesseract.exe") {
+            return $path
+        }
+    }
+    
+    return $null
+}
+
+# Find Tesseract path
+$tesseractPath = Find-TesseractPath
+
+if ($null -eq $tesseractPath) {
+    Write-Host "Tesseract OCR is not installed or was not found in standard locations." -ForegroundColor Red
+    Write-Host "Please install Tesseract OCR first from: https://github.com/UB-Mannheim/tesseract/wiki" -ForegroundColor Yellow
+    Write-Host "The script will close in 5 seconds..."
+    Start-Sleep -Seconds 5
+    exit
+}
+
+Write-Host "Tesseract OCR found at: $tesseractPath" -ForegroundColor Green
+
+# Check tessdata folder
+$tessDataPath = Join-Path -Path $tesseractPath -ChildPath "tessdata"
+if (-not (Test-Path $tessDataPath)) {
+    Write-Host "The tessdata folder doesn't exist. Creating..." -ForegroundColor Yellow
+    New-Item -Path $tessDataPath -ItemType Directory | Out-Null
+}
+
+# URLs of languages to download
+$languageUrls = @{
+    "fra" = "https://github.com/tesseract-ocr/tessdata/raw/4.00/fra.traineddata"
+    "fra_vert" = "https://github.com/tesseract-ocr/tessdata/raw/4.00/fra_vert.traineddata"
+    "frk" = "https://github.com/tesseract-ocr/tessdata/raw/4.00/frk.traineddata"
+    "frm" = "https://github.com/tesseract-ocr/tessdata/raw/4.00/frm.traineddata"
+}
+
+# Download languages
+foreach ($lang in $languageUrls.Keys) {
+    $url = $languageUrls[$lang]
+    $outputFile = Join-Path -Path $tessDataPath -ChildPath "$lang.traineddata"
+    
+    if (Test-Path $outputFile) {
+        Write-Host "Language file $lang already exists. Removing..." -ForegroundColor Yellow
+        Remove-Item -Path $outputFile -Force
+    }
+    
+    Write-Host "Downloading $lang.traineddata..." -ForegroundColor Cyan
+    
+    try {
+        Invoke-WebRequest -Uri $url -OutFile $outputFile
+        Write-Host "Language $lang successfully installed." -ForegroundColor Green
+    }
+    catch {
+        Write-Host "Error downloading $lang.traineddata: $_" -ForegroundColor Red
+    }
+}
+
+# Check installed languages
+Write-Host "`nVerifying installed languages:" -ForegroundColor Cyan
+$installedLanguages = Get-ChildItem -Path $tessDataPath -Filter "*.traineddata" | ForEach-Object { $_.Name.Replace(".traineddata", "") }
+Write-Host "Installed languages: $($installedLanguages -join ', ')" -ForegroundColor Green
+
+Write-Host "`nLanguage installation completed." -ForegroundColor Green
+Write-Host "Press any key to close..."
+$null = $Host.UI.RawUI.ReadKey("NoEcho,IncludeKeyDown") 

install_windows.bat

@@ -0,0 +1,61 @@
+@echo off
+echo ======================================================
+echo      Ragflow PDF Preprocessing Installation
+echo ======================================================
+echo.
+
+:: Check if Python is installed
+python --version > nul 2>&1
+if %ERRORLEVEL% NEQ 0 (
+    echo Error: Python is not installed or not in PATH.
+    echo Please install Python 3.8 or higher from https://www.python.org/downloads/
+    echo Make sure to check "Add Python to PATH" during installation.
+    pause
+    exit /b 1
+)
+
+:: Create virtual environment
+echo Creating virtual environment...
+python -m venv venv
+if %ERRORLEVEL% NEQ 0 (
+    echo Error creating virtual environment.
+    pause
+    exit /b 1
+)
+
+:: Activate virtual environment
+echo Activating virtual environment...
+call venv\Scripts\activate.bat
+
+:: Update pip
+echo Updating pip...
+python -m pip install --upgrade pip
+
+:: Install dependencies
+echo Installing dependencies...
+pip install -e .
+if %ERRORLEVEL% NEQ 0 (
+    echo Error installing dependencies.
+    pause
+    exit /b 1
+)
+
+echo.
+echo ======================================================
+echo Installation completed successfully!
+echo.
+echo Next steps:
+echo 1. Make sure Tesseract OCR is installed 
+echo    (https://github.com/UB-Mannheim/tesseract/wiki)
+echo 2. Make sure Ollama is installed and running
+echo    (https://ollama.ai/)
+echo 3. To launch the application, run:
+echo.
+echo    call venv\Scripts\activate.bat
+echo    python main.py
+echo.
+echo Or use the launch_windows.bat script
+echo ======================================================
+echo.
+
+pause 

install_windows.ps1

@@ -0,0 +1,107 @@
+# PowerShell installation script for Ragflow PDF Preprocessing
+Write-Host "======================================================" -ForegroundColor Cyan
+Write-Host "      Ragflow PDF Preprocessing Installation" -ForegroundColor Cyan
+Write-Host "======================================================" -ForegroundColor Cyan
+Write-Host ""
+
+# Check if PowerShell is running as administrator
+$isAdmin = ([Security.Principal.WindowsPrincipal] [Security.Principal.WindowsIdentity]::GetCurrent()).IsInRole([Security.Principal.WindowsBuiltInRole] "Administrator")
+if (-not $isAdmin) {
+    Write-Host "Note: For installing system dependencies, it is recommended to run this script as administrator." -ForegroundColor Yellow
+    Write-Host "You can continue without administrator privileges, but some features might not be installed correctly." -ForegroundColor Yellow
+    $continue = Read-Host "Do you want to continue? (Y/N)"
+    if ($continue -ne "Y" -and $continue -ne "y") {
+        Write-Host "Installation cancelled." -ForegroundColor Red
+        exit
+    }
+}
+
+# Check if Python is installed
+try {
+    $pythonVersion = python --version
+    Write-Host "Python detected: $pythonVersion" -ForegroundColor Green
+} catch {
+    Write-Host "Error: Python is not installed or not in PATH." -ForegroundColor Red
+    Write-Host "Please install Python 3.8 or higher from https://www.python.org/downloads/" -ForegroundColor Red
+    Write-Host "Make sure to check 'Add Python to PATH' during installation." -ForegroundColor Red
+    Read-Host "Press ENTER to exit"
+    exit
+}
+
+# Create virtual environment
+Write-Host "Creating virtual environment..." -ForegroundColor Cyan
+python -m venv venv
+if (-not $?) {
+    Write-Host "Error creating virtual environment." -ForegroundColor Red
+    Read-Host "Press ENTER to exit"
+    exit
+}
+
+# Activate virtual environment
+Write-Host "Activating virtual environment..." -ForegroundColor Cyan
+& .\venv\Scripts\Activate.ps1
+
+# Update pip
+Write-Host "Updating pip..." -ForegroundColor Cyan
+python -m pip install --upgrade pip
+
+# Install dependencies
+Write-Host "Installing dependencies..." -ForegroundColor Cyan
+pip install -e .
+if (-not $?) {
+    Write-Host "Error installing dependencies." -ForegroundColor Red
+    Read-Host "Press ENTER to exit"
+    exit
+}
+
+# Check if Tesseract is installed
+$tesseractPaths = @(
+    "C:\Program Files\Tesseract-OCR\tesseract.exe",
+    "C:\Program Files (x86)\Tesseract-OCR\tesseract.exe",
+    "C:\Tesseract-OCR\tesseract.exe"
+)
+
+$tesseractInstalled = $false
+foreach ($path in $tesseractPaths) {
+    if (Test-Path $path) {
+        $tesseractInstalled = $true
+        Write-Host "Tesseract OCR detected at: $path" -ForegroundColor Green
+        break
+    }
+}
+
+if (-not $tesseractInstalled) {
+    Write-Host "Tesseract OCR was not detected." -ForegroundColor Yellow
+    Write-Host "The application requires Tesseract OCR for text recognition." -ForegroundColor Yellow
+    Write-Host "Please install it from: https://github.com/UB-Mannheim/tesseract/wiki" -ForegroundColor Yellow
+}
+
+# Check if Ollama server is accessible
+try {
+    $response = Invoke-WebRequest -Uri "http://217.182.105.173:11434/api/version" -UseBasicParsing -ErrorAction SilentlyContinue
+    if ($response.StatusCode -eq 200) {
+        Write-Host "Ollama server is accessible at 217.182.105.173:11434." -ForegroundColor Green
+    }
+} catch {
+    Write-Host "Warning: Cannot connect to Ollama server at 217.182.105.173:11434." -ForegroundColor Yellow
+    Write-Host "Make sure you have network connectivity to the Ollama server." -ForegroundColor Yellow
+    Write-Host "This is required for the LLM features of the application." -ForegroundColor Yellow
+}
+
+Write-Host ""
+Write-Host "======================================================" -ForegroundColor Cyan
+Write-Host "Installation completed successfully!" -ForegroundColor Green
+Write-Host ""
+Write-Host "Next steps:" -ForegroundColor Cyan
+Write-Host "1. Make sure Tesseract OCR is installed" -ForegroundColor White
+Write-Host "   (https://github.com/UB-Mannheim/tesseract/wiki)" -ForegroundColor White
+Write-Host "2. Make sure you have network connectivity to the Ollama server at 217.182.105.173:11434" -ForegroundColor White
+Write-Host "3. To launch the application, run:" -ForegroundColor White
+Write-Host "" -ForegroundColor White
+Write-Host "   .\launch_windows.ps1" -ForegroundColor White
+Write-Host "" -ForegroundColor White
+Write-Host "Or use the launch_windows.bat script" -ForegroundColor White
+Write-Host "======================================================" -ForegroundColor Cyan
+Write-Host ""
+
+Read-Host "Press ENTER to exit" 

launch_windows.bat

@@ -0,0 +1,50 @@
+@echo off
+:: Script de lancement pour l'application de prétraitement PDF
+:: Ceci est un wrapper pour le script PowerShell qui lancer l'application sous Windows
+
+echo ===================================================
+echo  Lancement de l'application de prétraitement PDF
+echo ===================================================
+echo.
+
+:: Vérifier si PowerShell est disponible
+where powershell >nul 2>nul
+if %ERRORLEVEL% neq 0 (
+    echo ERREUR: PowerShell n'est pas disponible sur ce système.
+    echo Veuillez installer PowerShell pour exécuter cette application.
+    echo.
+    pause
+    exit /b 1
+)
+
+:: Récupérer le chemin du script
+set "SCRIPT_DIR=%~dp0"
+cd /d "%SCRIPT_DIR%"
+
+:: Vérifier si l'environnement virtuel existe
+if not exist "%SCRIPT_DIR%venv" (
+    echo L'environnement virtuel n'existe pas. Veuillez d'abord exécuter install_windows.bat
+    echo.
+    pause
+    exit /b 1
+)
+
+:: Définir l'exécution du script PowerShell pour permettre l'exécution non signée
+powershell -Command "Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass"
+
+:: Lancer le script PowerShell
+echo Lancement du script d'initialisation et de démarrage...
+powershell -ExecutionPolicy Bypass -File "%SCRIPT_DIR%launch_windows.ps1"
+
+:: Vérifier si le script s'est terminé correctement
+if %ERRORLEVEL% neq 0 (
+    echo Une erreur s'est produite lors du lancement de l'application.
+    echo Code d'erreur: %ERRORLEVEL%
+    echo.
+    echo Pour plus d'informations, veuillez vérifier les messages ci-dessus.
+    pause
+    exit /b %ERRORLEVEL%
+)
+
+:: Fin du script
+exit /b 0 

launch_windows.ps1

@@ -0,0 +1,84 @@
+# PowerShell launch script for Ragflow PDF Preprocessing
+Write-Host "======================================================" -ForegroundColor Cyan
+Write-Host "      Ragflow PDF Preprocessing Launch" -ForegroundColor Cyan
+Write-Host "======================================================" -ForegroundColor Cyan
+Write-Host ""
+
+# Répertoire actuel
+$scriptPath = Split-Path -Parent $MyInvocation.MyCommand.Path
+Set-Location $scriptPath
+
+# Vérifier si le répertoire venv existe
+if (-not (Test-Path "$scriptPath\venv")) {
+    Write-Host "L'environnement virtuel n'existe pas. Veuillez d'abord exécuter install_windows.ps1"
+    Write-Host "Vous pouvez double-cliquer sur 'install_windows.bat' pour lancer l'installation"
+    Read-Host "Appuyez sur Entrée pour quitter"
+    exit
+}
+
+# Activer l'environnement virtuel
+Write-Host "Activation de l'environnement virtuel..."
+& "$scriptPath\venv\Scripts\Activate.ps1"
+
+# Vérifier l'installation des dépendances
+Write-Host "Vérification des dépendances Python..."
+& python -c "import PyPDF2, PyQt6, pytesseract, Pillow" 2>$null
+if ($LASTEXITCODE -ne 0) {
+    Write-Host "Certaines dépendances ne sont pas installées. Installation en cours..."
+    & pip install -r requirements.txt
+}
+
+# Créer la structure de répertoires
+Write-Host "Initialisation de la structure de données..."
+& python -c "from utils.data_structure import initialize_data_directories; initialize_data_directories()"
+
+# Check if Tesseract is accessible
+$tesseractPaths = @(
+    "C:\Program Files\Tesseract-OCR\tesseract.exe",
+    "C:\Program Files (x86)\Tesseract-OCR\tesseract.exe",
+    "C:\Tesseract-OCR\tesseract.exe"
+)
+
+$tesseractInstalled = $false
+foreach ($path in $tesseractPaths) {
+    if (Test-Path $path) {
+        $tesseractInstalled = $true
+        Write-Host "Tesseract OCR detected at: $path" -ForegroundColor Green
+        break
+    }
+}
+
+if (-not $tesseractInstalled) {
+    Write-Host "Information: Tesseract OCR was not detected." -ForegroundColor Yellow
+    Write-Host "The application will try to find Tesseract in standard locations." -ForegroundColor Yellow
+    Write-Host "If OCR doesn't work, please install Tesseract from: https://github.com/UB-Mannheim/tesseract/wiki" -ForegroundColor Yellow
+}
+
+# Check if Ollama server is accessible
+try {
+    $response = Invoke-WebRequest -Uri "http://217.182.105.173:11434/api/version" -UseBasicParsing -ErrorAction SilentlyContinue
+    if ($response.StatusCode -eq 200) {
+        Write-Host "Ollama server is accessible at 217.182.105.173:11434." -ForegroundColor Green
+    }
+} catch {
+    Write-Host "Warning: Cannot connect to Ollama server at 217.182.105.173:11434." -ForegroundColor Yellow
+    Write-Host "Make sure you have network connectivity to the Ollama server." -ForegroundColor Yellow
+    $continue = Read-Host "Do you want to continue anyway? (Y/N)"
+    if ($continue -ne "Y" -and $continue -ne "y") {
+        exit
+    }
+}
+
+# Lancer l'application
+Write-Host "Lancement de l'application de prétraitement PDF..."
+& python main.py
+
+# End
+Write-Host ""
+if ($LASTEXITCODE -ne 0) {
+    Write-Host "The application has terminated with errors (code $LASTEXITCODE)." -ForegroundColor Red
+} else {
+    Write-Host "The application has terminated normally." -ForegroundColor Green
+}
+
+Read-Host "Press ENTER to exit" 

main.py

@@ -10,9 +10,13 @@ import os
 from PyQt6.QtWidgets import QApplication
 
 from ui.viewer import PDFViewer
+from utils.data_structure import initialize_data_directories
 
 def main():
     """Point d'entrée principal de l'application"""
+    # Initialiser la structure de données
+    initialize_data_directories()
+    
     app = QApplication(sys.argv)
     app.setApplicationName("Prétraitement PDF pour Ragflow")
     

ragflow_pretraitement.egg-info/PKG-INFO

@@ -0,0 +1,333 @@
+Metadata-Version: 2.4
+Name: ragflow_pretraitement
+Version: 1.0.0
+Summary: Outil de prétraitement PDF avec agents LLM modulables pour Ragflow
+Author: Ragflow Team
+Classifier: Programming Language :: Python :: 3
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Operating System :: OS Independent
+Requires-Python: >=3.8
+Description-Content-Type: text/markdown
+Requires-Dist: PyQt6>=6.4.0
+Requires-Dist: PyMuPDF>=1.21.0
+Requires-Dist: numpy>=1.22.0
+Requires-Dist: pytesseract>=0.3.9
+Requires-Dist: Pillow>=9.3.0
+Requires-Dist: opencv-python>=4.7.0
+Requires-Dist: requests>=2.28.0
+Dynamic: author
+Dynamic: classifier
+Dynamic: description
+Dynamic: description-content-type
+Dynamic: requires-dist
+Dynamic: requires-python
+Dynamic: summary
+
+# Prétraitement PDF pour Ragflow
+
+Outil de prétraitement de documents PDF avec agents LLM modulables pour l'analyse, la traduction et le résumé.
+
+## Fonctionnalités
+
+- **Sélection visuelle** de zones dans les documents PDF (tableaux, schémas, formules, texte)
+- **Analyse automatique** avec différents agents LLM configurables
+- **Niveaux d'analyse** adaptables selon les besoins (léger, moyen, avancé)
+- **Export Markdown** pour intégration dans une base Ragflow
+
+## Installation
+
+### Prérequis
+
+- Python 3.8 ou supérieur
+- PyQt6
+- PyMuPDF (fitz)
+- Tesseract OCR (pour la reconnaissance de texte)
+- Ollama (pour les modèles LLM)
+
+#### Vérification des prérequis (Windows)
+
+Un script de vérification est fourni pour vous aider à identifier les composants manquants:
+
+1. Faites un clic droit sur `check_prerequisites.ps1` et sélectionnez "Exécuter avec PowerShell"
+2. Le script vérifiera tous les prérequis et vous indiquera ce qu'il manque
+3. Suivez les instructions pour installer les composants manquants
+
+### Configuration de l'environnement virtuel
+
+Il est fortement recommandé d'utiliser un environnement virtuel pour installer et exécuter cette application, notamment sur Windows:
+
+#### Avec venv (recommandé)
+
+```bash
+# Windows (PowerShell)
+python -m venv venv
+.\venv\Scripts\Activate.ps1
+
+# Windows (CMD)
+python -m venv venv
+.\venv\Scripts\activate.bat
+
+# Linux/macOS
+python -m venv venv
+source venv/bin/activate
+```
+
+#### Avec Conda (alternative)
+
+```bash
+# Création de l'environnement
+conda create -n ragflow_env python=3.9
+conda activate ragflow_env
+
+# Installation des dépendances via pip
+pip install -r requirements.txt
+```
+
+Une fois l'environnement virtuel activé, vous pouvez procéder à l'installation.
+
+### Méthode 1 : Installation directe (recommandée)
+
+Clonez ce dépôt et installez avec pip :
+
+```bash
+git clone https://github.com/votre-utilisateur/ragflow-pretraitement.git
+cd ragflow_pretraitement
+pip install -e .
+```
+
+Cette méthode installera automatiquement toutes les dépendances Python requises.
+
+### Méthode 2 : Installation manuelle
+
+Si vous préférez une installation manuelle :
+
+```bash
+pip install -r requirements.txt
+```
+
+### Méthode 3 : Installation automatisée pour Windows
+
+Pour une installation simplifiée sous Windows, utilisez les scripts d'installation fournis :
+
+#### Avec les scripts batch (.bat)
+
+1. Double-cliquez sur `install_windows.bat`
+2. Le script créera un environnement virtuel et installera toutes les dépendances
+3. Suivez les instructions à l'écran pour les étapes suivantes
+
+Pour lancer l'application par la suite :
+1. Double-cliquez sur `launch_windows.bat`
+2. Le script vérifiera l'installation de Tesseract et Ollama avant de démarrer l'application
+
+#### Avec PowerShell (recommandé pour Windows 10/11)
+
+Si vous préférez utiliser PowerShell (interface plus conviviale avec code couleur) :
+
+1. Faites un clic droit sur `install_windows.ps1` et sélectionnez "Exécuter avec PowerShell"
+2. Suivez les instructions à l'écran pour terminer l'installation
+
+Pour lancer l'application par la suite :
+1. Faites un clic droit sur `launch_windows.ps1` et sélectionnez "Exécuter avec PowerShell"
+2. Le script effectuera des vérifications avant de démarrer l'application
+
+Ces méthodes sont recommandées pour les utilisateurs Windows qui ne sont pas familiers avec les lignes de commande.
+
+### Installation de Tesseract OCR
+
+Pour l'OCR, vous devez également installer Tesseract :
+
+- **Windows** : Téléchargez et installez depuis [https://github.com/UB-Mannheim/tesseract/wiki](https://github.com/UB-Mannheim/tesseract/wiki)
+  - L'application recherchera Tesseract dans les chemins standards sur Windows (`C:\Program Files\Tesseract-OCR\tesseract.exe`, `C:\Program Files (x86)\Tesseract-OCR\tesseract.exe` ou `C:\Tesseract-OCR\tesseract.exe`)
+  - Assurez-vous d'installer les langues françaises et anglaises pendant l'installation
+- **Linux** : `sudo apt install tesseract-ocr tesseract-ocr-fra tesseract-ocr-eng`
+- **macOS** : `brew install tesseract`
+
+### Connexion au serveur Ollama
+
+Cette application se connecte à un serveur Ollama distant pour les fonctionnalités LLM:
+
+- Adresse du serveur: `217.182.105.173:11434`
+- Modèles utilisés: 
+  - `mistral:latest` (pour le résumé et la traduction légère)
+  - `llava:34b-v1.6-fp16` (pour l'analyse visuelle)
+  - `llama3.2-vision:90b-instruct-q8_0` (pour l'analyse visuelle avancée)
+  - `deepseek-r1:70b-llama-distill-q8_0` (pour le résumé et la traduction avancés)
+  - `qwen2.5:72b-instruct-q8_0` (pour la traduction moyenne)
+
+Assurez-vous que votre machine dispose d'une connexion réseau vers ce serveur. Aucune installation locale d'Ollama n'est nécessaire.
+
+## Utilisation
+
+### Lancement de l'application
+
+Si vous avez utilisé l'installation avec le script setup.py :
+```bash
+ragflow-pretraitement
+```
+
+Ou manuellement :
+```bash
+cd ragflow_pretraitement
+python main.py
+```
+
+### Processus typique
+
+1. Charger un PDF avec le bouton "Charger PDF"
+2. Naviguer entre les pages avec les boutons ou le sélecteur
+3. Sélectionner une zone d'intérêt avec la souris
+4. Choisir le type de contenu (schéma, tableau, formule...)
+5. Ajouter un contexte textuel si nécessaire
+6. Configurer les agents LLM dans l'onglet "Agents LLM"
+7. Appliquer l'agent sur la sélection
+8. Exporter le résultat en Markdown
+
+## Configuration des agents LLM
+
+L'application propose trois niveaux d'analyse :
+
+| Niveau    | Vision                                | Résumé                        | Traduction                   | Usage recommandé            |
+| --------- | ------------------------------------- | ----------------------------- | ---------------------------- | --------------------------- |
+| 🔹 Léger  | llava:34b-v1.6-fp16                  | mistral:latest                | mistral:latest               | Débogage, prototypes       |
+| ⚪ Moyen   | llava:34b-v1.6-fp16                  | deepseek-r1:70b-llama-distill | qwen2.5:72b-instruct        | Usage normal               |
+| 🔸 Avancé | llama3.2-vision:90b-instruct-q8_0    | deepseek-r1:70b-llama-distill | deepseek-r1:70b-llama-distill| Documents critiques        |
+
+Vous pouvez personnaliser ces configurations dans le fichier `config/llm_profiles.json`.
+
+## Détail des bibliothèques utilisées
+
+L'application utilise les bibliothèques suivantes, chacune avec un rôle spécifique dans le traitement des documents:
+
+### Bibliothèques principales
+- **PyQt6** : Interface graphique complète (v6.4.0+)
+- **PyMuPDF (fitz)** : Manipulation et rendu des documents PDF (v1.21.0+)
+- **numpy** : Traitement numérique des images et des données (v1.22.0+)
+- **pytesseract** : Interface Python pour Tesseract OCR (v0.3.9+)
+- **Pillow** : Traitement d'images (v9.3.0+)
+- **opencv-python (cv2)** : Traitement d'images avancé et détection de contenu (v4.7.0+)
+- **requests** : Communication avec l'API Ollama (v2.28.0+)
+
+### Rôle de chaque bibliothèque
+- **PyQt6**: Framework d'interface graphique qui gère la visualisation PDF, les sélections interactives, la configuration des agents et l'interface utilisateur globale.
+- **PyMuPDF (fitz)**: Convertit les pages PDF en images, permet d'accéder au contenu des PDF, extraire les pages et obtenir des rendus haute qualité.
+- **numpy**: Manipule les données d'images sous forme de tableaux pour le traitement OCR et la détection de contenu.
+- **pytesseract**: Reconnaît le texte dans les images extraites des PDF, avec support multilingue.
+- **Pillow + opencv-python**: Prétraitement des images avant OCR pour améliorer la reconnaissance du texte.
+- **requests**: Envoie des requêtes au service Ollama local pour utiliser les modèles d'IA.
+
+## Structure du projet et modules clés
+
+```
+ragflow_pretraitement/
+├── main.py                   # Point d'entrée principal
+├── ui/                       # Interface utilisateur
+│   ├── viewer.py             # Visualisation PDF et sélection (PyQt6, fitz)
+│   └── llm_config_panel.py   # Configuration des agents (PyQt6)
+├── agents/                   # Agents LLM
+│   ├── base.py               # Classe de base des agents
+│   ├── vision.py             # Agent d'analyse visuelle (utilise OllamaAPI)
+│   ├── summary.py            # Agent de résumé (utilise OllamaAPI)
+│   ├── translation.py        # Agent de traduction (utilise OllamaAPI)
+│   └── rewriter.py           # Agent de reformulation (utilise OllamaAPI)
+├── utils/                    # Utilitaires
+│   ├── ocr.py                # Reconnaissance de texte (pytesseract, opencv)
+│   ├── markdown_export.py    # Export en Markdown
+│   └── api_ollama.py         # Communication avec l'API Ollama (requests)
+├── config/                   # Configuration
+│   └── llm_profiles.json     # Profils des modèles LLM utilisés
+└── data/                     # Données
+    └── outputs/              # Résultats générés
+```
+
+### Détail des modules principaux
+
+#### Interface utilisateur (ui/)
+- **viewer.py**: Implémente la classe principale `PDFViewer` qui gère la visualisation des PDF, la sélection des zones, le zoom et la navigation entre les pages.
+- **llm_config_panel.py**: Implémente la classe `LLMConfigPanel` qui permet de configurer les agents LLM, leurs paramètres et de lancer les traitements.
+
+#### Agents LLM (agents/)
+- **base.py**: Définit la classe de base `LLMBaseAgent` avec les méthodes communes à tous les agents.
+- **vision.py**: Agent spécialisé dans l'analyse d'images et de schémas.
+- **summary.py**: Agent pour résumer et synthétiser du texte.
+- **translation.py**: Agent pour traduire le contenu (généralement de l'anglais vers le français).
+- **rewriter.py**: Agent pour reformuler et améliorer du texte.
+
+#### Utilitaires (utils/)
+- **ocr.py**: Contient la classe `OCRProcessor` pour extraire du texte des images avec prétraitement avancé.
+- **api_ollama.py**: Implémente la classe `OllamaAPI` pour communiquer avec le service Ollama.
+- **markdown_export.py**: Gère l'export des résultats au format Markdown.
+
+## Adaptations pour Windows
+
+Cette version a été optimisée pour Windows avec les adaptations suivantes:
+
+1. **Chemins Tesseract**: Configuration automatique des chemins Tesseract pour Windows
+2. **URLs API**: Configuration de l'API Ollama pour utiliser `localhost` par défaut
+3. **Chemins de fichiers**: Utilisation de la séparation de chemins compatible Windows
+4. **Compatibilité Unicode**: Support des caractères spéciaux dans les chemins Windows
+
+## Fonctionnalités pratiques et cas d'utilisation
+
+### 1. Analyse de tableaux techniques
+- Sélectionnez un tableau complexe dans un document
+- Utilisez l'agent de vision pour reconnaître la structure
+- Exportez le contenu sous forme de markdown bien formaté
+
+### 2. Traduction de documents techniques
+- Chargez un document en anglais
+- Sélectionnez des sections de texte
+- Utilisez l'agent de traduction pour obtenir une version française précise
+- Exportez le résultat en markdown
+
+### 3. Résumé de documentation longue
+- Identifiez les sections importantes d'un long document
+- Appliquez l'agent de résumé à chaque section
+- Combinez les résultats en un document synthétique
+
+### 4. Analyse de schémas et figures
+- Sélectionnez un schéma complexe
+- Utilisez l'agent de vision pour obtenir une description détaillée
+- Ajoutez du contexte textuel au besoin
+
+## Résolution des problèmes courants
+
+### Erreurs d'importation de PyQt6
+
+Si vous rencontrez des erreurs avec PyQt6, essayez de le réinstaller :
+```bash
+pip uninstall PyQt6 PyQt6-Qt6 PyQt6-sip
+pip install PyQt6
+```
+
+### Problèmes avec l'OCR sur Windows
+
+Si l'OCR (Tesseract) ne fonctionne pas correctement :
+1. Vérifiez que Tesseract est correctement installé (redémarrez l'application après installation)
+2. Vérifiez l'existence d'un des chemins standards (`C:\Program Files\Tesseract-OCR\tesseract.exe`) 
+3. Si nécessaire, modifiez manuellement le chemin dans `utils/ocr.py`
+
+### Connectivité au serveur Ollama
+
+Si vous ne pouvez pas vous connecter au serveur Ollama:
+1. Vérifiez votre connexion réseau et assurez-vous que vous pouvez accéder à `217.182.105.173:11434`
+2. Vérifiez qu'aucun pare-feu ou proxy ne bloque la connexion
+3. Pour vérifier la disponibilité du serveur, essayez d'accéder à `http://217.182.105.173:11434/api/version` dans votre navigateur
+
+Si vous souhaitez utiliser une instance Ollama locale à la place:
+1. Modifiez l'URL dans `utils/api_ollama.py` et `agents/base.py` pour utiliser `http://localhost:11434`
+2. Installez Ollama localement depuis [https://ollama.ai/](https://ollama.ai/)
+3. Téléchargez les modèles requis avec `ollama pull <nom_du_modèle>`
+
+### Modèles manquants
+
+Si certains modèles ne fonctionnent pas, vérifiez leur disponibilité sur le serveur et modifiez si nécessaire les noms dans `config/llm_profiles.json`.
+
+## Limitations
+
+- OCR parfois imprécis sur les documents complexes ou de mauvaise qualité
+- Certains modèles nécessitent beaucoup de mémoire (8GB+ de RAM recommandé)
+- Les formules mathématiques complexes peuvent être mal interprétées
+
+## Contribution
+
+Les contributions sont les bienvenues ! N'hésitez pas à soumettre des pull requests ou à signaler des problèmes. 

ragflow_pretraitement.egg-info/SOURCES.txt

@@ -0,0 +1,21 @@
+README.md
+setup.py
+agents/__init__.py
+agents/base.py
+agents/rewriter.py
+agents/summary.py
+agents/translation.py
+agents/vision.py
+ragflow_pretraitement.egg-info/PKG-INFO
+ragflow_pretraitement.egg-info/SOURCES.txt
+ragflow_pretraitement.egg-info/dependency_links.txt
+ragflow_pretraitement.egg-info/entry_points.txt
+ragflow_pretraitement.egg-info/requires.txt
+ragflow_pretraitement.egg-info/top_level.txt
+ui/__init__.py
+ui/llm_config_panel.py
+ui/viewer.py
+utils/__init__.py
+utils/api_ollama.py
+utils/markdown_export.py
+utils/ocr.py

ragflow_pretraitement.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

ragflow_pretraitement.egg-info/entry_points.txt

@@ -0,0 +1,2 @@
+[console_scripts]
+ragflow-pretraitement = ragflow_pretraitement.main:main

ragflow_pretraitement.egg-info/requires.txt

@@ -0,0 +1,7 @@
+PyQt6>=6.4.0
+PyMuPDF>=1.21.0
+numpy>=1.22.0
+pytesseract>=0.3.9
+Pillow>=9.3.0
+opencv-python>=4.7.0
+requests>=2.28.0

ragflow_pretraitement.egg-info/top_level.txt

@@ -0,0 +1,3 @@
+agents
+ui
+utils

test_components.py

@@ -0,0 +1,234 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+
+"""
+Test script to verify critical components
+"""
+
+import os
+import sys
+import platform
+import requests
+import time
+import subprocess
+import json
+from typing import List, Dict, Any
+
+# Check Tesseract OCR installation
+def check_tesseract():
+    print("\n=== Checking Tesseract OCR ===")
+    try:
+        import pytesseract
+        from PIL import Image
+        
+        # Possible paths for Tesseract on Windows
+        possible_paths = [
+            r"C:\Program Files\Tesseract-OCR\tesseract.exe",
+            r"C:\Program Files (x86)\Tesseract-OCR\tesseract.exe",
+            r"C:\Tesseract-OCR\tesseract.exe",
+            r"C:\Users\PCDEV\AppData\Local\Programs\Tesseract-OCR\tesseract.exe",
+            r"C:\Users\PCDEV\Tesseract-OCR\tesseract.exe"
+        ]
+        
+        # Check if Tesseract is in PATH
+        tesseract_in_path = False
+        try:
+            if platform.system() == "Windows":
+                result = subprocess.run(["where", "tesseract"], capture_output=True, text=True)
+                if result.returncode == 0:
+                    tesseract_in_path = True
+                    tesseract_path = result.stdout.strip().split("\n")[0]
+                    print(f"Tesseract found in PATH: {tesseract_path}")
+            else:
+                result = subprocess.run(["which", "tesseract"], capture_output=True, text=True)
+                if result.returncode == 0:
+                    tesseract_in_path = True
+                    tesseract_path = result.stdout.strip()
+                    print(f"Tesseract found in PATH: {tesseract_path}")
+        except Exception as e:
+            print(f"Error checking for Tesseract in PATH: {e}")
+        
+        if not tesseract_in_path and platform.system() == "Windows":
+            print("Tesseract is not in PATH. Searching in standard locations...")
+            
+            # Check standard paths
+            for path in possible_paths:
+                if os.path.exists(path):
+                    pytesseract.pytesseract.tesseract_cmd = path
+                    print(f"Tesseract found at: {path}")
+                    break
+        
+        # Test Tesseract with a version command
+        try:
+            if platform.system() == "Windows" and not tesseract_in_path:
+                # Use explicit path
+                for path in possible_paths:
+                    if os.path.exists(path):
+                        result = subprocess.run([path, "--version"], capture_output=True, text=True)
+                        if result.returncode == 0:
+                            print(f"Tesseract version: {result.stdout.strip().split()[0]}")
+                            break
+            else:
+                # Tesseract is in PATH
+                result = subprocess.run(["tesseract", "--version"], capture_output=True, text=True)
+                if result.returncode == 0:
+                    print(f"Tesseract version: {result.stdout.strip().split()[0]}")
+        except Exception as e:
+            print(f"Error checking Tesseract version: {e}")
+        
+        # Check installed languages
+        try:
+            if platform.system() == "Windows" and not tesseract_in_path:
+                # Use explicit path
+                for path in possible_paths:
+                    if os.path.exists(path):
+                        tesseract_folder = os.path.dirname(path)
+                        tessdata_folder = os.path.join(tesseract_folder, "tessdata")
+                        if os.path.exists(tessdata_folder):
+                            langs = [f for f in os.listdir(tessdata_folder) if f.endswith(".traineddata")]
+                            print(f"Installed languages: {', '.join([lang.split('.')[0] for lang in langs])}")
+                            break
+            else:
+                # Tesseract is in PATH
+                result = subprocess.run(["tesseract", "--list-langs"], capture_output=True, text=True)
+                if result.returncode == 0:
+                    langs = result.stdout.strip().split("\n")[1:]  # Skip the first line
+                    print(f"Installed languages: {', '.join(langs)}")
+        except Exception as e:
+            print(f"Error checking Tesseract languages: {e}")
+        
+        print("\nINSTRUCTIONS FOR TESSERACT OCR:")
+        print("1. If Tesseract is not installed, download it from:")
+        print("   https://github.com/UB-Mannheim/tesseract/wiki")
+        print("2. Make sure to install French (fra) and English (eng) languages")
+        print("3. Check the 'Add to PATH' option during installation")
+        
+    except ImportError as e:
+        print(f"Error: {e}")
+        print("Tesseract OCR or its Python dependencies are not properly installed")
+        print("Install them with: pip install pytesseract Pillow")
+
+# Check connection to Ollama
+def check_ollama(endpoint="http://217.182.105.173:11434"):
+    print("\n=== Checking Ollama connection ===")
+    print(f"Endpoint: {endpoint}")
+    
+    # Test basic connection
+    try:
+        response = requests.get(f"{endpoint}/api/version", timeout=10)
+        if response.status_code == 200:
+            version_info = response.json()
+            print(f"✓ Connection to Ollama successful - Version: {version_info.get('version', 'unknown')}")
+            
+            # List available models
+            try:
+                response = requests.get(f"{endpoint}/api/tags", timeout=10)
+                if response.status_code == 200:
+                    models = response.json().get("models", [])
+                    if models:
+                        print(f"✓ Available models ({len(models)}):")
+                        for model in models:
+                            print(f"  - {model.get('name', 'Unknown')} ({model.get('size', 'Unknown size')})")
+                    else:
+                        print("No models found on Ollama server")
+                else:
+                    print(f"✗ Error retrieving models: status {response.status_code}")
+            except requests.exceptions.RequestException as e:
+                print(f"✗ Error retrieving models: {str(e)}")
+            
+            # Test a simple model
+            try:
+                print("\nTesting a simple model (mistral)...")
+                payload = {
+                    "model": "mistral",
+                    "prompt": "Say hello in English",
+                    "options": {
+                        "temperature": 0.1
+                    }
+                }
+                
+                start_time = time.time()
+                response = requests.post(f"{endpoint}/api/generate", json=payload, timeout=30)
+                elapsed_time = time.time() - start_time
+                
+                if response.status_code == 200:
+                    try:
+                        result = response.json()
+                        print(f"✓ Test successful in {elapsed_time:.2f} seconds")
+                        print(f"  Response: {result.get('response', 'No response')[:100]}...")
+                    except json.JSONDecodeError as e:
+                        print(f"✗ JSON parsing error: {str(e)}")
+                        print("  Trying to process first line only...")
+                        
+                        lines = response.text.strip().split("\n")
+                        if lines:
+                            try:
+                                result = json.loads(lines[0])
+                                print(f"✓ Test successful with first line parsing in {elapsed_time:.2f} seconds")
+                                print(f"  Response: {result.get('response', 'No response')[:100]}...")
+                            except json.JSONDecodeError:
+                                print("✗ Failed to parse first line as JSON")
+                                print(f"  Raw response (first 200 chars): {response.text[:200]}")
+                else:
+                    print(f"✗ Error testing model: status {response.status_code}")
+                    print(f"  Body: {response.text[:200]}")
+            except requests.exceptions.RequestException as e:
+                print(f"✗ Error testing model: {str(e)}")
+        else:
+            print(f"✗ Error connecting to Ollama: status {response.status_code}")
+            print(f"  Body: {response.text[:200]}")
+    except requests.exceptions.RequestException as e:
+        print(f"✗ Unable to connect to Ollama: {str(e)}")
+        
+    print("\nINSTRUCTIONS FOR OLLAMA:")
+    print("1. Verify that the Ollama server is running at the specified address")
+    print("2. Verify that port 11434 is open and accessible")
+    print("3. Check Ollama server logs for potential issues")
+
+# Check Python environment
+def check_python_env():
+    print("\n=== Checking Python environment ===")
+    print(f"Python {sys.version}")
+    print(f"Platform: {platform.platform()}")
+    
+    # Check installed packages
+    required_packages = ["PyQt6", "PyPDF2", "pytesseract", "requests", "fitz"]
+    
+    print("\nChecking required packages:")
+    for pkg in required_packages:
+        try:
+            __import__(pkg)
+            print(f"✓ {pkg} is installed")
+        except ImportError:
+            print(f"✗ {pkg} is NOT installed")
+    
+    # Check Pillow separately (package name is Pillow but import name is PIL)
+    try:
+        import PIL
+        print(f"✓ PIL (Pillow) is installed")
+    except ImportError:
+        print(f"✗ PIL (Pillow) is NOT installed")
+            
+    print("\nINSTRUCTIONS FOR PYTHON ENVIRONMENT:")
+    print("1. Make sure you're using the virtual environment if configured")
+    print("2. Install missing packages with: pip install -r requirements.txt")
+
+# Main function
+def main():
+    print("=== Testing critical components ===")
+    
+    # Check Python environment
+    check_python_env()
+    
+    # Check Tesseract OCR
+    check_tesseract()
+    
+    # Check connection to Ollama
+    check_ollama()
+    
+    print("\n=== Checks completed ===")
+    print("If issues were detected, follow the displayed instructions")
+    print("After fixing issues, run this script again to verify")
+
+if __name__ == "__main__":
+    main() 

ui/llm_config_panel.py

@@ -10,7 +10,8 @@ import json
 from PyQt6.QtWidgets import (QWidget, QVBoxLayout, QHBoxLayout, QLabel, 
                            QPushButton, QComboBox, QTextEdit, QGroupBox,
                            QListWidget, QSplitter, QTabWidget, QSpinBox,
-                           QDoubleSpinBox, QFormLayout, QMessageBox, QCheckBox)
+                           QDoubleSpinBox, QFormLayout, QMessageBox, QCheckBox,
+                           QApplication)
 from PyQt6.QtCore import Qt, QSize
 
 class LLMConfigPanel(QWidget):
@@ -121,6 +122,11 @@ class LLMConfigPanel(QWidget):
         self.summary_model_combo.addItems(["mistral", "deepseek-r1"])
         agent_type_layout.addRow("Modèle:", self.summary_model_combo)
         
+        # Case à cocher pour activer/désactiver l'agent de résumé
+        self.summary_enabled_checkbox = QCheckBox("Activer l'agent de résumé")
+        self.summary_enabled_checkbox.setChecked(False)  # Désactivé par défaut
+        agent_type_layout.addRow("", self.summary_enabled_checkbox)
+        
         # Agent de traduction
         agent_type_layout.addRow(QLabel("<b>Agent Traduction</b>"))
         
@@ -196,6 +202,11 @@ class LLMConfigPanel(QWidget):
         self.include_images_check.setChecked(True)
         export_group_layout.addWidget(self.include_images_check)
         
+        # Option pour enregistrer les images capturées
+        self.save_captured_images_check = QCheckBox("Enregistrer les images capturées")
+        self.save_captured_images_check.setChecked(True)
+        export_group_layout.addWidget(self.save_captured_images_check)
+        
         export_layout.addWidget(export_group)
         
         export_btn = QPushButton("Exporter")
@@ -338,51 +349,122 @@ class LLMConfigPanel(QWidget):
                 self.translation_model_combo.setCurrentText(translation["model"])
     
     def run_agent(self):
-        """Execute l'agent LLM sur la sélection actuelle"""
+        """Exécute l'agent LLM sur la sélection actuelle"""
         if not self.current_selection:
-            QMessageBox.warning(self, "Aucune sélection", 
-                                "Veuillez sélectionner une région à analyser.")
+            self.result_text.setText("Veuillez sélectionner une région avant d'exécuter l'agent.")
             return
         
-        # Pour cette démonstration, nous simulons le résultat
-        # Dans une implémentation réelle, nous utiliserions les agents LLM
+        try:
+            # Récupération des paramètres
+            selection_type = self.current_selection.get("type", "autre")
+            context = self.current_selection.get("context", "")
             
-        # Récupérer les paramètres
+            # Paramètres de génération
+            gen_params = {
+                "temperature": self.temp_spin.value(),
+                "top_p": self.top_p_spin.value(),
+                "max_tokens": self.token_spin.value(),
+                "save_images": self.save_captured_images_check.isChecked()  # Ajout du paramètre d'enregistrement
+            }
+            
+            self.result_text.setText("Analyse en cours...\nCette opération peut prendre plusieurs minutes.")
+            self.parent.status_bar.showMessage("Traitement de l'image en cours...")
+            
+            # Forcer la mise à jour de l'interface
+            QApplication.processEvents()
+            
+            # Afficher les informations sur la sélection actuelle (débogue)
+            rect = self.current_selection.get("rect")
+            if rect:
+                rect_info = f"Sélection: x={rect.x()}, y={rect.y()}, w={rect.width()}, h={rect.height()}, page={self.current_selection.get('page', 0)+1}"
+                print(f"INFO SÉLECTION: {rect_info}")
+            
+            # Récupérer l'image de la sélection
+            selection_image = self.parent.get_selection_image(self.current_selection)
+            
+            if selection_image:
+                # Vérifier la taille des données d'image pour le débogue
+                print(f"Taille des données image: {len(selection_image)} octets")
+                
+                # Récupération des paramètres depuis l'interface
                 vision_model = self.vision_model_combo.currentText()
+                vision_lang = self.vision_lang_combo.currentText()
                 summary_model = self.summary_model_combo.currentText()
                 translation_model = self.translation_model_combo.currentText()
-        lang = self.vision_lang_combo.currentText()
-        temp = self.temp_spin.value()
                 
-        # Exemple de résultat simulé
-        result = f"**Analyse de l'agent Vision ({vision_model})**\n\n"
+                # Configuration du pipeline de traitement
+                from utils.workflow_manager import WorkflowManager
+                from config.agent_config import ACTIVE_AGENTS
                 
-        if self.current_selection["type"] == "schéma":
-            result += "Le schéma illustre un processus en plusieurs étapes avec des connections entre les différents éléments.\n\n"
-        elif self.current_selection["type"] == "tableau":
-            result += "Le tableau contient des données structurées avec plusieurs colonnes et rangées.\n\n"
-        elif self.current_selection["type"] == "formule":
-            result += "La formule mathématique représente une équation complexe.\n\n"
-        else:
-            result += "Le contenu sélectionné a été analysé.\n\n"
+                # Mise à jour dynamique de l'activation des agents
+                active_agents = ACTIVE_AGENTS.copy()
+                active_agents["summary"] = self.summary_enabled_checkbox.isChecked()
                 
-        result += f"**Résumé ({summary_model})**\n\n"
-        result += "Ce contenu montre l'importance des éléments sélectionnés dans le contexte du document.\n\n"
+                # Afficher les agents actifs pour le débogage
+                print(f"Agents actifs: {active_agents}")
+                print(f"Modèles utilisés: vision={vision_model}, translation={translation_model}, summary={summary_model}")
                 
-        if lang == "en":
-            result += f"**Traduction ({translation_model})**\n\n"
-            result += "The selected content has been analyzed and shows the importance of the selected elements in the context of the document.\n\n"
+                # Créer le gestionnaire de workflow
+                workflow = WorkflowManager(
+                    vision_model=vision_model,
+                    summary_model=summary_model,
+                    translation_model=translation_model,
+                    active_agents=active_agents,
+                    config=gen_params
+                )
                 
-        result += "**Paramètres utilisés**\n"
-        result += f"modèle vision={vision_model}, modèle résumé={summary_model}, "
-        result += f"temperature={temp}, langue={lang}"
+                # Exécution du workflow
+                results = workflow.process_image(
+                    image_data=selection_image,
+                    content_type=selection_type,
+                    context=context,
+                    target_lang=vision_lang
+                )
                 
-        # Stocker et afficher le résultat
-        self.analysis_results[id(self.current_selection)] = result
-        self.result_text.setText(result)
+                # Affichage des résultats
+                if results:
+                    result_text = ""
                     
-        # Mise à jour du statut
+                    # Vision (original)
+                    if "vision" in results and results["vision"]:
+                        result_text += "🔍 ANALYSE VISUELLE (en):\n"
+                        result_text += results["vision"]
+                        result_text += "\n\n"
+                    
+                    # Traduction
+                    if "translation" in results and results["translation"]:
+                        result_text += "🌐 TRADUCTION (fr):\n" 
+                        result_text += results["translation"]
+                        result_text += "\n\n"
+                    
+                    # Résumé (si activé)
+                    if "summary" in results and results["summary"]:
+                        result_text += "📝 RÉSUMÉ (fr):\n"
+                        result_text += results["summary"]
+                    
+                    # Message d'erreur
+                    if "error" in results:
+                        result_text += "\n\n❌ ERREUR:\n"
+                        result_text += results["error"]
+                    
+                    # Enregistrer l'analyse dans l'historique
+                    self.analysis_results[self.selection_list.currentRow()] = results
+                    
+                    self.result_text.setText(result_text)
                     self.parent.status_bar.showMessage("Analyse terminée.")
+                else:
+                    self.result_text.setText("Erreur: Impossible d'obtenir une analyse de l'image.")
+                    self.parent.status_bar.showMessage("Analyse échouée.")
+            else:
+                self.result_text.setText("Erreur: Impossible d'extraire l'image de la sélection.")
+                self.parent.status_bar.showMessage("Erreur: Extraction d'image impossible.")
+                
+        except Exception as e:
+            import traceback
+            error_details = traceback.format_exc()
+            print(f"Erreur lors de l'analyse: {error_details}")
+            self.result_text.setText(f"Erreur lors de l'analyse: {str(e)}\n\nDétails de l'erreur:\n{error_details}")
+            self.parent.status_bar.showMessage("Erreur: Analyse échouée.")
     
     def export_results(self):
         """Exporte les résultats au format Markdown"""

ui/viewer.py

@@ -164,7 +164,12 @@ class PDFViewer(QMainWindow):
         zoom_matrix = fitz.Matrix(2 * self.zoom_factor, 2 * self.zoom_factor)
         
         # Rendu de la page en un pixmap
+        try:
+            # Essayer avec la nouvelle API (PyMuPDF récent)
             pixmap = page.get_pixmap(matrix=zoom_matrix, alpha=False)
+        except AttributeError:
+            # Fallback pour les anciennes versions
+            pixmap = page.render_pixmap(matrix=zoom_matrix, alpha=False)
         
         # Conversion en QImage puis QPixmap
         img = QImage(pixmap.samples, pixmap.width, pixmap.height, 
@@ -224,14 +229,31 @@ class PDFViewer(QMainWindow):
         """
         page = self.current_page if page_num is None else page_num
         
-        # Ajuster les coordonnées selon le zoom actuel
+        # DEBUG: Afficher les coordonnées brutes et le facteur de zoom
+        print(f"DEBUG: Coordonnées brutes de sélection: x={rect.x()}, y={rect.y()}, w={rect.width()}, h={rect.height()}")
+        print(f"DEBUG: Facteur de zoom: {self.zoom_factor}")
+        
+        # Calculer le décalage du viewport
+        viewport_pos = self.scroll_area.horizontalScrollBar().value(), self.scroll_area.verticalScrollBar().value()
+        print(f"DEBUG: Décalage du viewport: x={viewport_pos[0]}, y={viewport_pos[1]}")
+        
+        # Obtenir les dimensions actuelles de l'image affichée
+        if self.pdf_label.pixmap():
+            img_width = self.pdf_label.pixmap().width()
+            img_height = self.pdf_label.pixmap().height()
+            print(f"DEBUG: Dimensions de l'image affichée: {img_width}x{img_height}")
+        
+        # Ajuster les coordonnées en fonction du zoom et du décalage
+        # Note: pour simplifier, nous ignorons le décalage du viewport pour l'instant
         adjusted_rect = QRectF(
-            rect.x() / self.zoom_factor,
-            rect.y() / self.zoom_factor,
-            rect.width() / self.zoom_factor,
-            rect.height() / self.zoom_factor
+            rect.x() / self.zoom_factor / 2,  # Division par 2 car la matrice de rendu utilise 2x
+            rect.y() / self.zoom_factor / 2,
+            rect.width() / self.zoom_factor / 2,
+            rect.height() / self.zoom_factor / 2
         )
         
+        print(f"DEBUG: Coordonnées ajustées: x={adjusted_rect.x()}, y={adjusted_rect.y()}, w={adjusted_rect.width()}, h={adjusted_rect.height()}")
+        
         selection = {
             "page": page,
             "rect": adjusted_rect,
@@ -245,6 +267,63 @@ class PDFViewer(QMainWindow):
         # Informer l'utilisateur
         self.status_bar.showMessage(f"Sélection ajoutée à la page {page + 1}")
 
+    def get_selection_image(self, selection):
+        """
+        Extrait l'image de la sélection
+        
+        Args:
+            selection (dict): Dictionnaire contenant les informations de la sélection
+            
+        Returns:
+            bytes: Données de l'image en bytes, ou None en cas d'erreur
+        """
+        if not selection or not self.pdf_document:
+            return None
+            
+        try:
+            # Récupérer la page
+            page_num = selection.get("page", 0)
+            page = self.pdf_document[page_num]
+            
+            # Récupérer les coordonnées de la sélection
+            rect = selection.get("rect")
+            if not rect:
+                return None
+                
+            # Convertir les coordonnées en rectangle PyMuPDF
+            # Les coordonnées stockées sont déjà ajustées par rapport au zoom (voir méthode add_selection)
+            mupdf_rect = fitz.Rect(
+                rect.x(), 
+                rect.y(), 
+                rect.x() + rect.width(), 
+                rect.y() + rect.height()
+            )
+            
+            # Debug: Afficher les coordonnées pour vérification
+            print(f"Coordonnées de la sélection (ajustées): {mupdf_rect}")
+            
+            # Matrice de transformation pour la qualité (sans zoom supplémentaire)
+            matrix = fitz.Matrix(2, 2)  # Facteur de qualité fixe à 2
+            
+            # Capturer l'image de la zone sélectionnée
+            try:
+                # Essayer avec la nouvelle API (PyMuPDF récent)
+                pix = page.get_pixmap(matrix=matrix, clip=mupdf_rect)
+            except AttributeError:
+                # Fallback pour les anciennes versions
+                pix = page.render_pixmap(matrix=matrix, clip=mupdf_rect)
+            
+            # Debug: Afficher les dimensions du pixmap obtenu
+            print(f"Dimensions de l'image capturée: {pix.width}x{pix.height}")
+            
+            # Convertir en format PNG
+            img_data = pix.tobytes("png")
+            
+            return img_data
+            
+        except Exception as e:
+            print(f"Erreur lors de l'extraction de l'image: {str(e)}")
+            return None
 
 class PDFLabel(QLabel):
     """Étiquette personnalisée pour afficher le PDF et gérer les sélections"""

utils/api_ollama.py

@@ -2,86 +2,254 @@
 # -*- coding: utf-8 -*-
 
 """
-Module pour l'interaction avec l'API Ollama
+API Interface for Ollama
 """
 
-import json
 import requests
+import json
 import base64
+import time
+import os
+import threading
 from typing import List, Dict, Any, Optional, Union, Callable
 
+# Verrouillage global pour les appels Ollama
+_ollama_lock = threading.Lock()
+_model_in_use = None
+_last_call_time = 0.0  # Float pour le temps (secondes)
+_min_delay_between_calls = 3.0  # Délai minimum en secondes entre les appels à Ollama
 
 class OllamaAPI:
     """
-    Classe pour interagir avec l'API Ollama
+    Simplified interface for Ollama API
     """
     
     def __init__(self, base_url: str = "http://217.182.105.173:11434"):
         """
-        Initialise la connexion à l'API Ollama
+        Initialize the API with the server's base URL
         
         Args:
-            base_url (str): URL de base de l'API Ollama
+            base_url (str): Base URL of the Ollama server
         """
-        self.base_url = base_url.rstrip("/")
+        self.base_url = base_url
         self.generate_endpoint = f"{self.base_url}/api/generate"
         self.chat_endpoint = f"{self.base_url}/api/chat"
         self.models_endpoint = f"{self.base_url}/api/tags"
+        self.timeout = 120  # Increase timeout to 2 minutes
+        self.max_retries = 2  # Nombre maximum de tentatives pour les requêtes
+        self.retry_delay = 2  # Délai entre les tentatives en secondes
         
-    def list_models(self) -> List[str]:
+        # Check connection on startup
+        self._check_connection()
+    
+    @staticmethod
+    def wait_for_ollama(model_name: str, timeout: int = 120) -> bool:
         """
-        Récupère la liste des modèles disponibles
+        Attend que le serveur Ollama soit disponible pour le modèle spécifié
+        
+        Args:
+            model_name (str): Nom du modèle à attendre
+            timeout (int): Délai d'attente maximum en secondes
             
         Returns:
-            List[str]: Liste des noms de modèles disponibles
+            bool: True si le serveur est disponible, False si timeout
+        """
+        global _ollama_lock, _model_in_use, _last_call_time, _min_delay_between_calls
+        
+        # Calculer le temps à attendre depuis le dernier appel
+        time_since_last_call = time.time() - _last_call_time
+        if time_since_last_call < _min_delay_between_calls:
+            delay = _min_delay_between_calls - time_since_last_call
+            print(f"Attente de {delay:.1f}s pour respecter le délai minimal entre appels...")
+            time.sleep(delay)
+        
+        start_time = time.time()
+        
+        while True:
+            with _ollama_lock:
+                # Si aucun modèle n'est en cours d'utilisation ou si c'est le modèle demandé
+                if _model_in_use is None:
+                    _model_in_use = model_name
+                    _last_call_time = time.time()
+                    return True
+                
+                # Si le temps d'attente est dépassé
+                if time.time() - start_time > timeout:
+                    print(f"Timeout en attendant Ollama pour le modèle {model_name}")
+                    return False
+            
+            # Attendre et réessayer
+            wait_time = min(5, (timeout - (time.time() - start_time)))
+            if wait_time <= 0:
+                return False
+                
+            print(f"En attente d'Ollama ({_model_in_use} est en cours d'utilisation)... Nouvel essai dans {wait_time:.1f}s")
+            time.sleep(wait_time)
+    
+    @staticmethod
+    def release_ollama():
+        """Libère le verrouillage sur Ollama"""
+        global _ollama_lock, _model_in_use, _last_call_time
+        
+        with _ollama_lock:
+            _model_in_use = None
+            _last_call_time = time.time()
+            print("Ollama libéré et disponible pour de nouveaux appels")
+    
+    def _check_connection(self) -> bool:
+        """
+        Checks if the Ollama server is accessible
+        
+        Returns:
+            bool: True if server is accessible, False otherwise
         """
         try:
-            response = requests.get(self.models_endpoint)
-            response.raise_for_status()
-            data = response.json()
+            response = requests.get(f"{self.base_url}/api/version", timeout=10)
+            if response.status_code == 200:
+                version_info = response.json()
+                print(f"Connection to Ollama established. Version: {version_info.get('version', 'unknown')}")
+                return True
+            else:
+                print(f"Error connecting to Ollama: status {response.status_code}")
+                return False
+        except requests.exceptions.RequestException as e:
+            print(f"Unable to connect to Ollama server: {str(e)}")
+            print(f"URL: {self.base_url}")
+            print("Check that the server is running and accessible.")
+            return False
     
-            # Extraire les noms des modèles
-            models = [model['name'] for model in data.get('models', [])]
-            return models
+    def list_models(self) -> List[Dict[str, Any]]:
+        """
+        Lists available models on Ollama server
         
-        except Exception as e:
-            print(f"Erreur lors de la récupération des modèles: {str(e)}")
+        Returns:
+            List[Dict[str, Any]]: List of available models
+        """
+        try:
+            response = requests.get(self.models_endpoint, timeout=self.timeout)
+            if response.status_code == 200:
+                return response.json().get("models", [])
+            else:
+                print(f"Error retrieving models: status {response.status_code}")
                 return []
+        except requests.exceptions.RequestException as e:
+            print(f"Connection error while retrieving models: {str(e)}")
+            return []
+    
+    def _is_model_available(self, model_name: str) -> bool:
+        """
+        Vérifie si un modèle spécifique est disponible sur le serveur
+        
+        Args:
+            model_name (str): Nom du modèle à vérifier
+            
+        Returns:
+            bool: True si le modèle est disponible, False sinon
+        """
+        models = self.list_models()
+        available_models = [model["name"] for model in models]
+        
+        # Vérification exacte
+        if model_name in available_models:
+            return True
+        
+        # Vérification partielle (pour gérer les versions)
+        for available_model in available_models:
+            # Si le modèle demandé est une partie d'un modèle disponible
+            if model_name in available_model or available_model in model_name:
+                print(f"Note: Le modèle '{model_name}' correspond partiellement à '{available_model}'")
+                return True
+        
+        return False
+    
+    def _make_request_with_retry(self, method: str, url: str, json_data: Dict[str, Any], 
+                                timeout: Optional[int] = None) -> requests.Response:
+        """
+        Effectue une requête HTTP avec mécanisme de réessai
+        
+        Args:
+            method (str): Méthode HTTP (POST, GET, etc.)
+            url (str): URL de la requête
+            json_data (Dict): Données JSON à envoyer
+            timeout (int, optional): Timeout en secondes
+            
+        Returns:
+            requests.Response: Réponse HTTP
+            
+        Raises:
+            requests.exceptions.RequestException: Si toutes les tentatives échouent
+        """
+        # Utiliser la valeur par défaut de l'instance si aucun timeout n'est spécifié
+        request_timeout = self.timeout if timeout is None else timeout
+            
+        attempt = 0
+        last_error = None
+        
+        while attempt < self.max_retries:
+            try:
+                if method.upper() == "POST":
+                    return requests.post(url, json=json_data, timeout=request_timeout)
+                elif method.upper() == "GET":
+                    return requests.get(url, json=json_data, timeout=request_timeout)
+                else:
+                    raise ValueError(f"Méthode HTTP non supportée: {method}")
+            except requests.exceptions.RequestException as e:
+                last_error = e
+                attempt += 1
+                if attempt < self.max_retries:
+                    print(f"Tentative {attempt} échouée. Nouvelle tentative dans {self.retry_delay}s...")
+                    time.sleep(self.retry_delay)
+        
+        # Si on arrive ici, c'est que toutes les tentatives ont échoué
+        raise last_error or requests.exceptions.RequestException("Toutes les tentatives ont échoué")
     
     def generate(self, model: str, prompt: str, images: Optional[List[bytes]] = None, 
                  options: Optional[Dict[str, Any]] = None) -> Dict[str, Any]:
         """
-        Génère une réponse à partir d'un prompt
+        Generates a response from an Ollama model
         
         Args:
-            model (str): Nom du modèle à utiliser
-            prompt (str): Texte du prompt
-            images (List[bytes], optional): Liste d'images en bytes
-            options (Dict, optional): Options de génération
+            model (str): Model name to use
+            prompt (str): Prompt text
+            images (List[bytes], optional): Images to send to model (for multimodal models)
+            options (Dict, optional): Generation options
             
         Returns:
-            Dict[str, Any]: Réponse du modèle
+            Dict[str, Any]: Model response
         """
-        # Options par défaut
-        default_options = {
-            "temperature": 0.2,
-            "top_p": 0.95,
-            "top_k": 40,
-            "num_predict": 1024
-        }
+        # Default response in case of errors
+        result: Dict[str, Any] = {"error": "Unknown error", "response": "Error during generation"}
         
-        # Fusionner avec les options fournies
-        if options:
-            default_options.update(options)
+        # Input validation
+        if not model:
+            return {"error": "Model parameter is required", "response": "Error: no model specified"}
             
-        # Construire la payload
+        if not prompt and not images:
+            return {"error": "Either prompt or images must be provided", "response": "Error: no content to generate from"}
+        
+        if options is None:
+            options = {}
+        
+        # Vérifier si le modèle est disponible
+        if not self._is_model_available(model):
+            model_error = f"Le modèle '{model}' n'est pas disponible sur le serveur Ollama. Utilisez la commande: ollama pull {model}"
+            print(model_error)
+            return {"error": model_error, "response": f"Error: model '{model}' not found, try pulling it first"}
+        
+        # Attendre que le serveur Ollama soit disponible
+        if not self.wait_for_ollama(model, timeout=180):
+            return {"error": "Timeout waiting for Ollama", "response": "Timeout waiting for Ollama server to be available"}
+        
+        try:
+            # Prepare payload
             payload = {
                 "model": model,
                 "prompt": prompt,
-            "options": default_options
+                "options": options,
+                "stream": False  # Important: disable streaming to avoid JSON parsing errors
             }
             
-        # Ajouter les images si fournies (pour les modèles multimodaux)
+            # Add images if provided (for multimodal models)
             if images:
                 base64_images = []
                 for img in images:
@@ -91,56 +259,114 @@ class OllamaAPI:
                 
                 payload["images"] = base64_images
             
-        try:
-            # Envoyer la requête
-            response = requests.post(self.generate_endpoint, json=payload)
-            response.raise_for_status()
-            return response.json()
+            # Make request
+            print(f"Sending request to {self.generate_endpoint} for model {model}...")
+            start_time = time.time()
             
-        except requests.exceptions.HTTPError as e:
-            print(f"Erreur HTTP: {e}")
-            return {"error": str(e)}
+            try:
+                response = self._make_request_with_retry("POST", self.generate_endpoint, payload)
+            except requests.exceptions.RequestException as e:
+                self.release_ollama()  # Libérer Ollama en cas d'erreur
+                return {"error": f"Connection error: {str(e)}", "response": "Error connecting to model server"}
+                
+            elapsed_time = time.time() - start_time
+            
+            # Handle response
+            if response.status_code == 200:
+                print(f"Response received in {elapsed_time:.2f} seconds")
+                try:
+                    result = response.json()
+                except Exception as e:
+                    # In case of JSON parsing error, try to process line by line
+                    print(f"JSON parsing error: {e}")
+                    print("Trying to process line by line...")
+                    
+                    # If the response contains multiple JSON lines, take the first valid line
+                    lines = response.text.strip().split("\n")
+                    if len(lines) > 0:
+                        try:
+                            result = json.loads(lines[0])
+                        except:
+                            # If that still doesn't work, return the raw text
+                            result = {"response": response.text[:1000], "model": model}
+            elif response.status_code == 404:
+                # Modèle spécifiquement non trouvé
+                error_msg = f"Model '{model}' not found on the server. Try running: ollama pull {model}"
+                print(error_msg)
+                result = {"error": error_msg, "response": f"Error: model '{model}' not found, try pulling it first"}
+            else:
+                error_msg = f"Error during generation: status {response.status_code}"
+                try:
+                    error_json = response.json()
+                    if "error" in error_json:
+                        error_msg += f", message: {error_json['error']}"
+                except:
+                    error_msg += f", body: {response.text[:100]}"
+                
+                print(error_msg)
+                result = {"error": error_msg, "response": "Error communicating with model"}
         
         except Exception as e:
-            print(f"Erreur lors de la génération: {str(e)}")
-            return {"error": str(e)}
+            # Catch any other unexpected errors
+            error_msg = f"Unexpected error: {str(e)}"
+            print(error_msg)
+            result = {"error": error_msg, "response": "An unexpected error occurred"}
+            
+        finally:
+            # Toujours libérer Ollama à la fin
+            self.release_ollama()
+        
+        # Ensure we always return a dictionary
+        return result
     
     def chat(self, model: str, messages: List[Dict[str, Any]], 
              images: Optional[List[bytes]] = None,
              options: Optional[Dict[str, Any]] = None) -> Dict[str, Any]:
         """
-        Utilise l'API de chat pour une conversation
+        Generates a response from a chat history
         
         Args:
-            model (str): Nom du modèle à utiliser
-            messages (List[Dict]): Liste des messages de la conversation
-                Format: [{"role": "user", "content": "message"}, ...]
-            images (List[bytes], optional): Liste d'images en bytes (pour le dernier message)
-            options (Dict, optional): Options de génération
+            model (str): Model name to use
+            messages (List[Dict]): List of chat messages (format [{"role": "user", "content": "..."}])
+            images (List[bytes], optional): Images to send to model (for multimodal models)
+            options (Dict, optional): Generation options
             
         Returns:
-            Dict[str, Any]: Réponse du modèle
+            Dict[str, Any]: Model response
         """
-        # Options par défaut
-        default_options = {
-            "temperature": 0.2,
-            "top_p": 0.95,
-            "top_k": 40,
-            "num_predict": 1024
-        }
+        # Default response in case of errors
+        result: Dict[str, Any] = {"error": "Unknown error", "response": "Error during chat generation"}
         
-        # Fusionner avec les options fournies
-        if options:
-            default_options.update(options)
+        # Input validation
+        if not model:
+            return {"error": "Model parameter is required", "response": "Error: no model specified"}
             
-        # Construire la payload
+        if not messages:
+            return {"error": "Messages parameter is required", "response": "Error: no chat messages provided"}
+        
+        if options is None:
+            options = {}
+        
+        # Vérifier si le modèle est disponible
+        if not self._is_model_available(model):
+            model_error = f"Le modèle '{model}' n'est pas disponible sur le serveur Ollama. Utilisez la commande: ollama pull {model}"
+            print(model_error)
+            return {"error": model_error, "response": f"Error: model '{model}' not found, try pulling it first"}
+        
+        # Attendre que le serveur Ollama soit disponible
+        if not self.wait_for_ollama(model, timeout=180):
+            return {"error": "Timeout waiting for Ollama", "response": "Timeout waiting for Ollama server to be available"}
+        
+        try:
+            # Prepare payload
             payload = {
                 "model": model,
                 "messages": messages,
-            "options": default_options
+                "options": options,
+                "stream": False  # Important: disable streaming to avoid JSON parsing errors
             }
             
-        # Ajouter les images au dernier message utilisateur si fournies
+            # Add images to the last user message if provided
             if images and messages and messages[-1]["role"] == "user":
                 base64_images = []
                 for img in images:
@@ -148,97 +374,161 @@ class OllamaAPI:
                         base64_img = base64.b64encode(img).decode("utf-8")
                         base64_images.append(base64_img)
                 
-            # Modifier le dernier message pour inclure les images
+                # Modify the last message to include images
                 last_message = messages[-1].copy()
-            # Les images doivent être dans un champ distinct du modèle d'API d'Ollama
-            # Pas comme un champ texte standard mais dans un tableau d'images
-            if "images" not in last_message:
                 last_message["images"] = base64_images
                 
-            # Remplacer le dernier message
+                # Replace the last message
                 payload["messages"] = messages[:-1] + [last_message]
             
-        try:
-            # Envoyer la requête
-            response = requests.post(self.chat_endpoint, json=payload)
-            response.raise_for_status()
-            return response.json()
+            # Make request
+            print(f"Sending chat request to {self.chat_endpoint} for model {model}...")
+            start_time = time.time()
             
-        except requests.exceptions.HTTPError as e:
-            print(f"Erreur HTTP: {e}")
-            return {"error": str(e)}
+            try:
+                response = self._make_request_with_retry("POST", self.chat_endpoint, payload)
+            except requests.exceptions.RequestException as e:
+                self.release_ollama()  # Libérer Ollama en cas d'erreur
+                return {"error": f"Connection error: {str(e)}", "response": "Error connecting to model server"}
+                
+            elapsed_time = time.time() - start_time
+            
+            # Handle response
+            if response.status_code == 200:
+                print(f"Chat response received in {elapsed_time:.2f} seconds")
+                try:
+                    result = response.json()
+                except Exception as e:
+                    # In case of JSON parsing error, try to process line by line
+                    print(f"JSON parsing error: {e}")
+                    lines = response.text.strip().split("\n")
+                    if len(lines) > 0:
+                        try:
+                            result = json.loads(lines[0])
+                        except:
+                            result = {"message": {"content": response.text[:1000]}, "model": model}
+            elif response.status_code == 404:
+                # Modèle spécifiquement non trouvé
+                error_msg = f"Model '{model}' not found on the server. Try running: ollama pull {model}"
+                print(error_msg)
+                result = {"error": error_msg, "response": f"Error: model '{model}' not found, try pulling it first"}
+            else:
+                error_msg = f"Error during chat generation: status {response.status_code}"
+                try:
+                    error_json = response.json()
+                    if "error" in error_json:
+                        error_msg += f", message: {error_json['error']}"
+                except:
+                    error_msg += f", body: {response.text[:100]}"
+                
+                print(error_msg)
+                result = {"error": error_msg, "response": "Error communicating with model"}
         
         except Exception as e:
-            print(f"Erreur lors du chat: {str(e)}")
-            return {"error": str(e)}
+            # Catch any other unexpected errors
+            error_msg = f"Unexpected error: {str(e)}"
+            print(error_msg)
+            result = {"error": error_msg, "response": "An unexpected error occurred"}
+            
+        finally:
+            # Toujours libérer Ollama à la fin
+            self.release_ollama()
+        
+        # Ensure we always return a dictionary
+        return result
     
     def stream_generate(self, model: str, prompt: str, 
                         callback: Callable[[str], None],
-                       options: Optional[Dict[str, Any]] = None) -> None:
+                        images: Optional[List[bytes]] = None,
+                        options: Optional[Dict[str, Any]] = None) -> str:
         """
-        Génère une réponse en streaming et appelle le callback pour chaque morceau
+        Generate a response in streaming mode with a callback function
         
         Args:
-            model (str): Nom du modèle à utiliser
-            prompt (str): Texte du prompt
-            callback (Callable): Fonction à appeler pour chaque morceau de texte
-            options (Dict, optional): Options de génération
+            model (str): Model name
+            prompt (str): Prompt to send
+            callback (Callable): Function called for each received chunk
+            images (List[bytes], optional): Images to send
+            options (Dict, optional): Generation options
+            
+        Returns:
+            str: Complete generated text
         """
-        # Options par défaut
-        default_options = {
-            "temperature": 0.2,
-            "top_p": 0.95,
-            "top_k": 40,
-            "num_predict": 1024,
-            "stream": True  # Activer le streaming
-        }
+        if options is None:
+            options = {}
         
-        # Fusionner avec les options fournies
-        if options:
-            default_options.update(options)
-            
-        # S'assurer que stream est activé
-        default_options["stream"] = True
-        
-        # Construire la payload
         payload = {
             "model": model,
             "prompt": prompt,
-            "options": default_options
+            "options": options,
+            "stream": True  # Enable streaming
         }
         
-        try:
-            # Envoyer la requête en streaming
-            with requests.post(self.generate_endpoint, json=payload, stream=True) as response:
-                response.raise_for_status()
+        # Add images if provided
+        if images:
+            base64_images = []
+            for img in images:
+                if isinstance(img, bytes):
+                    base64_img = base64.b64encode(img).decode("utf-8")
+                    base64_images.append(base64_img)
+            
+            payload["images"] = base64_images
+        
+        full_response = ""
+        
+        try:
+            with requests.post(
+                self.generate_endpoint,
+                json=payload,
+                stream=True,
+                timeout=self.timeout
+            ) as response:
+                if response.status_code != 200:
+                    error_msg = f"Error during streaming: status {response.status_code}"
+                    callback(error_msg)
+                    return error_msg
                 
-                # Traiter chaque ligne de la réponse
                 for line in response.iter_lines():
                     if line:
                         try:
-                            data = json.loads(line)
-                            if "response" in data:
-                                callback(data["response"])
+                            chunk = json.loads(line)
+                            if "response" in chunk:
+                                text_chunk = chunk["response"]
+                                full_response += text_chunk
+                                callback(text_chunk)
                         except json.JSONDecodeError:
-                            print(f"Erreur de décodage JSON: {line}")
+                            # Ignore lines that are not valid JSON
+                            pass
                 
-        except requests.exceptions.HTTPError as e:
-            print(f"Erreur HTTP: {e}")
-            callback(f"\nErreur: {str(e)}")
+            return full_response
             
         except Exception as e:
-            print(f"Erreur lors du streaming: {str(e)}")
-            callback(f"\nErreur: {str(e)}")
+            error_msg = f"Error during streaming: {str(e)}"
+            callback(error_msg)
+            return error_msg
 
-    def check_connection(self) -> bool:
-        """
-        Vérifie si la connexion à l'API Ollama est fonctionnelle
+# Test the API if executed directly
+if __name__ == "__main__":
+    api = OllamaAPI()
+    print("Testing connection to Ollama...")
     
-        Returns:
-            bool: True si la connexion est établie, False sinon
-        """
-        try:
-            response = requests.get(f"{self.base_url}/api/version")
-            return response.status_code == 200
-        except:
-            return False 
+    if api._check_connection():
+        print("Connection successful!")
+        
+        print("\nList of available models:")
+        models = api.list_models()
+        for model in models:
+            print(f"- {model.get('name', 'Unknown')} ({model.get('size', 'Unknown size')})")
+        
+        print("\nTesting a model (if available):")
+        if models and "name" in models[0]:
+            model_name = models[0]["name"]
+            print(f"Testing model {model_name} with a simple prompt...")
+            response = api.generate(model_name, "Say hello in English")
+            if "response" in response:
+                print(f"Response: {response['response']}")
+            else:
+                print(f"Error: {response.get('error', 'Unknown error')}")
+    else:
+        print("Failed to connect to Ollama.")
+        print("Check that the server is running at the specified address.") 

utils/data_structure.py

@@ -0,0 +1,65 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+
+"""
+Utilitaire pour initialiser la structure de données du projet
+"""
+
+import os
+import platform
+
+def initialize_data_directories():
+    """
+    Crée la structure de répertoires nécessaire pour le stockage des données
+    """
+    # Répertoire principal de données
+    data_dir = "data"
+    
+    # Sous-répertoires pour les différents types de données
+    subdirs = [
+        "outputs",           # Sorties finales
+        "images",            # Images analysées
+        "translations",      # Journaux de traduction
+        "ocr_logs",          # Journaux OCR
+        "summaries",         # Journaux de résumés
+        "rewrites",          # Journaux de reformulation
+        "uploads",           # Fichiers téléchargés temporaires
+        "templates",         # Modèles de documents
+        "temp",              # Fichiers temporaires
+        "workflows"          # Journaux de flux de travail complets
+    ]
+    
+    # Créer le répertoire principal s'il n'existe pas
+    if not os.path.exists(data_dir):
+        os.makedirs(data_dir)
+        print(f"Création du répertoire principal: {data_dir}")
+    
+    # Créer les sous-répertoires
+    for subdir in subdirs:
+        full_path = os.path.join(data_dir, subdir)
+        if not os.path.exists(full_path):
+            os.makedirs(full_path)
+            print(f"Création du sous-répertoire: {full_path}")
+    
+    # Créer le fichier .gitkeep dans chaque répertoire vide pour le contrôle de version
+    for subdir in subdirs:
+        full_path = os.path.join(data_dir, subdir)
+        gitkeep_file = os.path.join(full_path, ".gitkeep")
+        if not os.path.exists(gitkeep_file):
+            with open(gitkeep_file, "w") as f:
+                f.write("# Ce fichier permet de conserver le répertoire vide dans Git")
+    
+    print("Structure de répertoires initialisée avec succès.")
+    
+    # Ajustements spécifiques à Windows
+    if platform.system() == "Windows":
+        # Vérifier si le chemin du fichier temporaire est trop long
+        temp_dir = os.path.join(data_dir, "temp")
+        temp_path = os.path.abspath(temp_dir)
+        if len(temp_path) > 240:  # Windows a une limite de 260 caractères pour les chemins
+            print(f"AVERTISSEMENT: Le chemin du répertoire temporaire est trop long: {len(temp_path)} caractères")
+            print("Windows a une limite de 260 caractères pour les chemins complets.")
+            print("Considérez déplacer le projet dans un répertoire avec un chemin plus court.")
+
+if __name__ == "__main__":
+    initialize_data_directories() 

utils/ocr.py

@@ -12,11 +12,23 @@ from typing import Union, Dict, Tuple, Optional
 from PIL import Image
 import io
 import copy
+import os
 
 # Configuration du chemin de Tesseract OCR
-# pytesseract.pytesseract.tesseract_cmd = r'C:\Program Files\Tesseract-OCR\tesseract.exe'  # Pour Windows
+pytesseract.pytesseract.tesseract_cmd = r'C:\Program Files\Tesseract-OCR\tesseract.exe'  # Pour Windows
 # Pour Linux et macOS, Tesseract doit être installé et disponible dans le PATH
 
+# Vérification si le chemin de Tesseract existe, sinon essayer d'autres chemins courants
+if not os.path.exists(pytesseract.pytesseract.tesseract_cmd):
+    alternative_paths = [
+        r'C:\Program Files (x86)\Tesseract-OCR\tesseract.exe',
+        r'C:\Tesseract-OCR\tesseract.exe'
+    ]
+    for path in alternative_paths:
+        if os.path.exists(path):
+            pytesseract.pytesseract.tesseract_cmd = path
+            break
+
 class OCRProcessor:
     """
     Classe pour traiter les images et extraire le texte

utils/pdf_processor.py

@@ -0,0 +1,138 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+
+"""
+Traitement des PDF avec le gestionnaire de flux de travail
+"""
+
+import os
+import io
+import time
+from typing import Dict, Any, Optional, List, Tuple, Union
+from PIL import Image
+
+from utils.workflow_manager import WorkflowManager
+
+class PDFProcessor:
+    """
+    Processeur de PDF intégré avec le gestionnaire de flux de travail
+    """
+    
+    def __init__(self):
+        """
+        Initialise le processeur de PDF
+        """
+        # Initialiser le gestionnaire de flux
+        self.workflow_manager = WorkflowManager()
+        
+        # Dossier pour sauvegarder les résultats
+        self.output_dir = os.path.join("data", "outputs")
+        os.makedirs(self.output_dir, exist_ok=True)
+        
+        # Timestamp pour cette session
+        self.timestamp = time.strftime("%Y%m%d-%H%M%S")
+        
+        print(f"Processeur PDF initialisé")
+    
+    def process_image_selection(self, image_data: bytes, selection_type: str, 
+                               context: str, page_number: int) -> Dict[str, Any]:
+        """
+        Traite une sélection d'image dans un PDF
+        
+        Args:
+            image_data (bytes): Données de l'image sélectionnée
+            selection_type (str): Type de sélection (schéma, tableau, etc.)
+            context (str): Contexte textuel associé à l'image
+            page_number (int): Numéro de la page où se trouve l'image
+            
+        Returns:
+            Dict: Résultats du traitement
+        """
+        print(f"Traitement d'une image de type '{selection_type}' (page {page_number})")
+        
+        # Traiter l'image avec le gestionnaire de flux
+        results = self.workflow_manager.process_image_with_context(
+            image_data, selection_type, context
+        )
+        
+        # Ajouter des informations supplémentaires aux résultats
+        results["page_number"] = page_number
+        results["timestamp"] = time.strftime("%Y-%m-%d %H:%M:%S")
+        
+        # Sauvegarder les résultats dans un fichier texte
+        self._save_results(results, selection_type, page_number)
+        
+        return results
+    
+    def _save_results(self, results: Dict[str, Any], selection_type: str, page_number: int) -> str:
+        """
+        Sauvegarde les résultats du traitement dans un fichier
+        
+        Args:
+            results (Dict): Résultats du traitement
+            selection_type (str): Type de sélection
+            page_number (int): Numéro de page
+            
+        Returns:
+            str: Chemin du fichier de résultats
+        """
+        # Créer un nom de fichier unique
+        filename = f"{self.timestamp}_page{page_number}_{selection_type}.txt"
+        output_path = os.path.join(self.output_dir, filename)
+        
+        # Sauvegarder l'image si disponible
+        if "original" in results and "image" in results["original"]:
+            try:
+                image_data = results["original"]["image"]
+                img = Image.open(io.BytesIO(image_data))
+                
+                image_filename = f"{self.timestamp}_page{page_number}_{selection_type}.png"
+                image_path = os.path.join(self.output_dir, image_filename)
+                
+                img.save(image_path)
+                print(f"Image sauvegardée dans: {image_path}")
+            except Exception as e:
+                print(f"Erreur lors de la sauvegarde de l'image: {str(e)}")
+        
+        # Écrire les résultats dans un fichier texte
+        with open(output_path, "w", encoding="utf-8") as f:
+            # En-tête
+            f.write(f"# Analyse de {selection_type} - Page {page_number}\n")
+            f.write(f"Date et heure: {results['timestamp']}\n\n")
+            
+            # Contexte original
+            if "original" in results and "context" in results["original"]:
+                f.write("## Contexte original (français)\n\n")
+                f.write(f"{results['original']['context']}\n\n")
+            
+            # Contexte traduit
+            if "context_en" in results:
+                f.write("## Contexte traduit (anglais)\n\n")
+                f.write(f"{results['context_en']}\n\n")
+            
+            # Analyse de vision
+            if "vision_analysis" in results:
+                f.write("## Analyse de l'image (anglais)\n\n")
+                f.write(f"{results['vision_analysis']}\n\n")
+            
+            # Analyse traduite
+            if "analysis_fr" in results:
+                f.write("## Analyse traduite (français)\n\n")
+                f.write(f"{results['analysis_fr']}\n\n")
+            
+            # Erreurs éventuelles
+            if "error" in results:
+                f.write("## ERREUR\n\n")
+                f.write(f"{results['error']}\n\n")
+        
+        print(f"Résultats sauvegardés dans: {output_path}")
+        return output_path
+    
+    def get_workflow_log_path(self) -> str:
+        """
+        Renvoie le chemin du fichier de journal du flux de travail
+        
+        Returns:
+            str: Chemin du fichier de journal
+        """
+        return self.workflow_manager.logger.get_workflow_path() 

utils/workflow_logger.py

@@ -0,0 +1,147 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+
+"""
+Utilitaire pour enregistrer le flux de traitement entre les agents LLM
+"""
+
+import os
+import time
+import uuid
+import json
+from typing import Dict, Any, Optional, List
+
+class WorkflowLogger:
+    """
+    Enregistreur de flux de travail pour suivre le traitement entre les agents
+    """
+    
+    def __init__(self, session_id: Optional[str] = None):
+        """
+        Initialise l'enregistreur de flux de travail
+        
+        Args:
+            session_id (str, optional): Identifiant de session (généré si non fourni)
+        """
+        # Utiliser l'ID fourni ou en générer un nouveau
+        self.session_id = session_id or str(uuid.uuid4())[:8]
+        self.timestamp = time.strftime("%Y%m%d-%H%M%S")
+        
+        # Répertoire pour les journaux
+        self.log_dir = os.path.join("data", "workflows")
+        os.makedirs(self.log_dir, exist_ok=True)
+        
+        # Fichier principal pour ce flux de travail
+        self.workflow_file = os.path.join(self.log_dir, f"{self.timestamp}_{self.session_id}_workflow.md")
+        
+        # Initialiser le fichier avec un en-tête
+        with open(self.workflow_file, "w", encoding="utf-8") as f:
+            f.write(f"# Flux de traitement - Session {self.session_id}\n\n")
+            f.write(f"Date et heure: {time.strftime('%d/%m/%Y %H:%M:%S')}\n\n")
+            f.write("## Étapes du traitement\n\n")
+    
+    def log_step(self, agent_name: str, input_data: Dict[str, Any], output_data: Any, 
+                 metadata: Optional[Dict[str, Any]] = None) -> None:
+        """
+        Enregistre une étape du flux de traitement
+        
+        Args:
+            agent_name (str): Nom de l'agent utilisé
+            input_data (Dict): Données d'entrée de l'agent
+            output_data (Any): Données de sortie de l'agent
+            metadata (Dict, optional): Métadonnées supplémentaires
+        """
+        step_time = time.strftime("%H:%M:%S")
+        
+        with open(self.workflow_file, "a", encoding="utf-8") as f:
+            # En-tête de l'étape
+            f.write(f"### Étape: {agent_name} ({step_time})\n\n")
+            
+            # Métadonnées si fournies
+            if metadata:
+                f.write("#### Métadonnées\n\n")
+                for key, value in metadata.items():
+                    f.write(f"- **{key}**: {value}\n")
+                f.write("\n")
+            
+            # Entrées
+            f.write("#### Entrées\n\n")
+            for key, value in input_data.items():
+                if key == "images" and value:
+                    f.write(f"- **{key}**: {len(value)} image(s) fournie(s)\n")
+                else:
+                    # Limiter la taille des entrées affichées
+                    if isinstance(value, str) and len(value) > 500:
+                        preview = value[:497] + "..."
+                        f.write(f"- **{key}**:\n```\n{preview}\n```\n")
+                    else:
+                        f.write(f"- **{key}**:\n```\n{value}\n```\n")
+            
+            # Sorties
+            f.write("#### Sorties\n\n")
+            if isinstance(output_data, str) and len(output_data) > 1000:
+                preview = output_data[:997] + "..."
+                f.write(f"```\n{preview}\n```\n")
+            else:
+                f.write(f"```\n{output_data}\n```\n")
+            
+            # Séparateur
+            f.write("\n---\n\n")
+    
+    def log_error(self, agent_name: str, error_message: str, 
+                  input_data: Optional[Dict[str, Any]] = None) -> None:
+        """
+        Enregistre une erreur dans le flux de traitement
+        
+        Args:
+            agent_name (str): Nom de l'agent qui a généré l'erreur
+            error_message (str): Message d'erreur
+            input_data (Dict, optional): Données d'entrée qui ont causé l'erreur
+        """
+        step_time = time.strftime("%H:%M:%S")
+        
+        with open(self.workflow_file, "a", encoding="utf-8") as f:
+            # En-tête de l'erreur
+            f.write(f"### ERREUR dans {agent_name} ({step_time})\n\n")
+            
+            # Message d'erreur
+            f.write("#### Message d'erreur\n\n")
+            f.write(f"```\n{error_message}\n```\n\n")
+            
+            # Entrées si fournies
+            if input_data:
+                f.write("#### Entrées ayant causé l'erreur\n\n")
+                for key, value in input_data.items():
+                    if key == "images" and value:
+                        f.write(f"- **{key}**: {len(value)} image(s) fournie(s)\n")
+                    else:
+                        # Limiter la taille des entrées affichées
+                        if isinstance(value, str) and len(value) > 500:
+                            preview = value[:497] + "..."
+                            f.write(f"- **{key}**:\n```\n{preview}\n```\n")
+                        else:
+                            f.write(f"- **{key}**:\n```\n{value}\n```\n")
+            
+            # Séparateur
+            f.write("\n---\n\n")
+    
+    def log_summary(self, summary_text: str) -> None:
+        """
+        Ajoute un résumé final au flux de traitement
+        
+        Args:
+            summary_text (str): Texte du résumé
+        """
+        with open(self.workflow_file, "a", encoding="utf-8") as f:
+            f.write("## Résumé du traitement\n\n")
+            f.write(f"{summary_text}\n\n")
+            f.write(f"*Fin du flux de traitement - {time.strftime('%d/%m/%Y %H:%M:%S')}*\n")
+    
+    def get_workflow_path(self) -> str:
+        """
+        Renvoie le chemin du fichier de flux de travail
+        
+        Returns:
+            str: Chemin du fichier
+        """
+        return self.workflow_file 

utils/workflow_manager.py

@@ -0,0 +1,320 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+
+"""
+Gestionnaire de flux de travail pour coordonner les agents LLM
+"""
+
+import os
+import time
+from typing import Dict, Any, Optional, List, Tuple, Callable, Type, Union
+
+from agents.ocr import OCRAgent
+from agents.vision import VisionAgent
+from agents.translation import TranslationAgent
+from agents.summary import SummaryAgent
+from agents.rewriter import RewriterAgent
+from agents.base import LLMBaseAgent
+
+from config.agent_config import (
+    is_agent_enabled, 
+    get_default_model, 
+    DEFAULT_ENDPOINT,
+    VERBOSE_LOGGING,
+    ACTIVE_AGENTS
+)
+
+from utils.workflow_logger import WorkflowLogger
+
+class WorkflowManager:
+    """
+    Gestionnaire de flux de travail pour coordonner les agents
+    """
+    
+    def __init__(self, vision_model: Optional[str] = None, summary_model: Optional[str] = None, 
+                 translation_model: Optional[str] = None, active_agents: Optional[Dict[str, bool]] = None,
+                 config: Optional[Dict[str, Any]] = None):
+        """
+        Initialise le gestionnaire de flux de travail
+        
+        Args:
+            vision_model (str, optional): Modèle à utiliser pour l'agent vision
+            summary_model (str, optional): Modèle à utiliser pour l'agent résumé
+            translation_model (str, optional): Modèle à utiliser pour l'agent traduction
+            active_agents (Dict[str, bool], optional): Dictionnaire des agents actifs
+            config (Dict[str, Any], optional): Configuration supplémentaire
+        """
+        # Créer un ID unique pour cette session
+        self.session_id = f"session_{int(time.time())}"
+        
+        # Initialiser le logger de flux
+        self.logger = WorkflowLogger(self.session_id)
+        
+        # Stocker les modèles spécifiés
+        self.model_overrides = {
+            "vision": vision_model,
+            "summary": summary_model,
+            "translation": translation_model
+        }
+        
+        # Stocker la configuration
+        self.config = config or {}
+        
+        # Utiliser les agents actifs spécifiés ou la configuration par défaut
+        self.active_agents = active_agents if active_agents is not None else ACTIVE_AGENTS.copy()
+        
+        # Initialiser les agents actifs selon la configuration
+        self.agents = {}
+        self._init_agents()
+        
+        print(f"Gestionnaire de flux initialisé avec session ID: {self.session_id}")
+        print(f"Agents actifs: {', '.join(self.agents.keys())}")
+    
+    def _init_agents(self) -> None:
+        """
+        Initialise les agents selon la configuration
+        """
+        # OCR Agent
+        if self.active_agents.get("ocr", True):
+            self.agents["ocr"] = OCRAgent(model_name="ocr", endpoint=DEFAULT_ENDPOINT)
+        
+        # Vision Agent (analyse d'images)
+        if self.active_agents.get("vision", True):
+            model = self.model_overrides["vision"] or get_default_model("vision")
+            self.agents["vision"] = VisionAgent(model, DEFAULT_ENDPOINT)
+        
+        # Translation Agent
+        if self.active_agents.get("translation", True):
+            model = self.model_overrides["translation"] or get_default_model("translation")
+            self.agents["translation"] = TranslationAgent(model, DEFAULT_ENDPOINT)
+        
+        # Summary Agent (désactivé selon la configuration)
+        if self.active_agents.get("summary", False):
+            model = self.model_overrides["summary"] or get_default_model("summary")
+            self.agents["summary"] = SummaryAgent(model, DEFAULT_ENDPOINT)
+        
+        # Rewriter Agent
+        if self.active_agents.get("rewriter", True):
+            model = get_default_model("rewriter")
+            self.agents["rewriter"] = RewriterAgent(model, DEFAULT_ENDPOINT)
+    
+    def process_image(self, image_data: bytes, content_type: str, 
+                      context: str = "", target_lang: str = "fr") -> Dict[str, Any]:
+        """
+        Traite une image avec contexte en utilisant les agents configurés
+        
+        Args:
+            image_data (bytes): Données de l'image
+            content_type (str): Type de contenu (schéma, tableau, etc.)
+            context (str): Contexte textuel associé à l'image
+            target_lang (str): Langue cible pour les résultats (fr par défaut)
+            
+        Returns:
+            Dict: Résultats du traitement
+        """
+        results = {}
+        
+        # Vérifier si l'option d'enregistrement des images est désactivée
+        save_images = self.config.get("save_images", True)
+        if not save_images and "vision" in self.agents:
+            # Appliquer la configuration au modèle de vision
+            self.agents["vision"].config["save_images"] = False
+        
+        try:
+            # Étape 1: Analyser l'image avec le contexte
+            if "vision" in self.agents:
+                print(f"Étape 1/3: Analyse de l'image avec le modèle {self.agents['vision'].model_name}...")
+                
+                # Si le contexte est en français et qu'on utilise un modèle en anglais,
+                # on pourrait le traduire ici, mais pour simplifier on l'utilise tel quel
+                vision_result = self.agents["vision"].generate(
+                    images=[image_data],
+                    selection_type=content_type,
+                    context=context
+                )
+                results["vision"] = vision_result
+                
+                # Journalisation
+                self.logger.log_step(
+                    agent_name="vision",
+                    input_data={"content_type": content_type, "context": context},
+                    output_data=vision_result,
+                    metadata={"model": self.agents["vision"].model_name}
+                )
+                
+                # Si nous avons une erreur ou un message vide, ne pas continuer
+                if "Error" in vision_result or "Erreur" in vision_result or not vision_result.strip():
+                    print(f"Avertissement: L'analyse d'image a échoué ou retourné un résultat vide.")
+                    if not "translation" in self.agents:
+                        return results
+                
+                # Attendre un peu avant le prochain appel pour éviter de surcharger le serveur
+                print("Délai de sécurité entre les appels d'agents (2s)...")
+                time.sleep(2)
+            
+            # Étape 2: Traduire l'analyse en français si nécessaire
+            if "translation" in self.agents and target_lang == "fr":
+                text_to_translate = results.get("vision", "")
+                
+                # Si nous n'avons pas de résultat de vision mais avons un contexte, utiliser le contexte
+                if (not text_to_translate or "Error" in text_to_translate or "Erreur" in text_to_translate) and context:
+                    text_to_translate = f"Voici le contexte de l'image: {context}"
+                elif not text_to_translate:
+                    text_to_translate = "Aucune analyse d'image disponible."
+                
+                print(f"Étape 2/3: Traduction du texte avec le modèle {self.agents['translation'].model_name}...")
+                
+                translation_result = self.agents["translation"].generate(
+                    prompt=text_to_translate,
+                    source_language="en",
+                    target_language="fr"
+                )
+                results["translation"] = translation_result
+                
+                # Journalisation
+                self.logger.log_step(
+                    agent_name="translation",
+                    input_data={"text": text_to_translate},
+                    output_data=translation_result,
+                    metadata={"model": self.agents["translation"].model_name}
+                )
+                
+                # Attendre un peu avant le prochain appel
+                print("Délai de sécurité entre les appels d'agents (2s)...")
+                time.sleep(2)
+            
+            # Étape 3: Générer un résumé si l'agent est activé
+            if "summary" in self.agents:
+                # Déterminer le texte à résumer
+                text_to_summarize = ""
+                if "translation" in results:
+                    text_to_summarize = results["translation"]
+                elif "vision" in results:
+                    text_to_summarize = results["vision"]
+                else:
+                    text_to_summarize = context or "Aucun texte disponible pour le résumé."
+                
+                if text_to_summarize.strip():
+                    print(f"Étape 3/3: Génération du résumé avec le modèle {self.agents['summary'].model_name}...")
+                    
+                    # Utiliser le texte traduit pour le résumé
+                    summary_result = self.agents["summary"].generate(
+                        prompt=text_to_summarize,
+                        language="fr" if "translation" in results else "en"
+                    )
+                    results["summary"] = summary_result
+                    
+                    # Journalisation
+                    self.logger.log_step(
+                        agent_name="summary",
+                        input_data={"text": text_to_summarize},
+                        output_data=summary_result,
+                        metadata={"model": self.agents["summary"].model_name}
+                    )
+            
+            # Générer un résumé du flux de travail
+            models_used = {
+                agent_name: agent.model_name
+                for agent_name, agent in self.agents.items()
+                if agent_name in ["vision", "translation", "summary"]
+            }
+            
+            summary = f"Traitement terminé avec succès.\n"
+            summary += f"Modèles utilisés: {', '.join([f'{k}={v}' for k, v in models_used.items()])}\n"
+            summary += f"Agents exécutés: {', '.join(results.keys())}"
+            
+            self.logger.log_summary(summary)
+            print(f"Traitement terminé avec succès.")
+            
+            return results
+            
+        except Exception as e:
+            error_msg = f"Erreur lors du traitement de l'image: {str(e)}"
+            print(error_msg)
+            self.logger.log_error("workflow_manager", error_msg)
+            results["error"] = error_msg
+            return results
+    
+    def _translate_text(self, text: str, source_lang: str, target_lang: str) -> str:
+        """
+        Traduit un texte d'une langue à une autre
+        
+        Args:
+            text (str): Texte à traduire
+            source_lang (str): Langue source
+            target_lang (str): Langue cible
+            
+        Returns:
+            str: Texte traduit
+        """
+        if not text.strip():
+            return ""
+        
+        # Préparer les entrées pour le logger
+        input_data = {
+            "prompt": text,
+            "source_language": source_lang,
+            "target_language": target_lang
+        }
+        
+        # Traduire le texte
+        translation_agent = self.agents["translation"]
+        translated_text = translation_agent.generate(
+            prompt=text,
+            source_language=source_lang,
+            target_language=target_lang
+        )
+        
+        # Enregistrer l'étape
+        self.logger.log_step(
+            agent_name="translation",
+            input_data=input_data,
+            output_data=translated_text,
+            metadata={
+                "source_language": source_lang,
+                "target_language": target_lang,
+                "model": translation_agent.model_name
+            }
+        )
+        
+        return translated_text
+    
+    def _analyze_image(self, image_data: bytes, selection_type: str, context: str) -> str:
+        """
+        Analyse une image avec contexte à l'aide de l'agent de vision
+        
+        Args:
+            image_data (bytes): Données de l'image
+            selection_type (str): Type de sélection
+            context (str): Contexte en anglais
+            
+        Returns:
+            str: Analyse de l'image
+        """
+        # Préparer les entrées pour le logger
+        input_data = {
+            "images": [image_data],
+            "selection_type": selection_type,
+            "context": context
+        }
+        
+        # Analyser l'image
+        vision_agent = self.agents["vision"]
+        analysis = vision_agent.generate(
+            images=[image_data],
+            selection_type=selection_type,
+            context=context
+        )
+        
+        # Enregistrer l'étape
+        self.logger.log_step(
+            agent_name="vision",
+            input_data=input_data,
+            output_data=analysis,
+            metadata={
+                "selection_type": selection_type,
+                "model": vision_agent.model_name
+            }
+        )
+        
+        return analysis 

venv/Lib/site-packages/PIL/BdfFontFile.py

@@ -0,0 +1,133 @@
+#
+# The Python Imaging Library
+# $Id$
+#
+# bitmap distribution font (bdf) file parser
+#
+# history:
+# 1996-05-16 fl   created (as bdf2pil)
+# 1997-08-25 fl   converted to FontFile driver
+# 2001-05-25 fl   removed bogus __init__ call
+# 2002-11-20 fl   robustification (from Kevin Cazabon, Dmitry Vasiliev)
+# 2003-04-22 fl   more robustification (from Graham Dumpleton)
+#
+# Copyright (c) 1997-2003 by Secret Labs AB.
+# Copyright (c) 1997-2003 by Fredrik Lundh.
+#
+# See the README file for information on usage and redistribution.
+#
+
+"""
+Parse X Bitmap Distribution Format (BDF)
+"""
+from __future__ import annotations
+
+from typing import BinaryIO
+
+from . import FontFile, Image
+
+bdf_slant = {
+    "R": "Roman",
+    "I": "Italic",
+    "O": "Oblique",
+    "RI": "Reverse Italic",
+    "RO": "Reverse Oblique",
+    "OT": "Other",
+}
+
+bdf_spacing = {"P": "Proportional", "M": "Monospaced", "C": "Cell"}
+
+
+def bdf_char(
+    f: BinaryIO,
+) -> (
+    tuple[
+        str,
+        int,
+        tuple[tuple[int, int], tuple[int, int, int, int], tuple[int, int, int, int]],
+        Image.Image,
+    ]
+    | None
+):
+    # skip to STARTCHAR
+    while True:
+        s = f.readline()
+        if not s:
+            return None
+        if s[:9] == b"STARTCHAR":
+            break
+    id = s[9:].strip().decode("ascii")
+
+    # load symbol properties
+    props = {}
+    while True:
+        s = f.readline()
+        if not s or s[:6] == b"BITMAP":
+            break
+        i = s.find(b" ")
+        props[s[:i].decode("ascii")] = s[i + 1 : -1].decode("ascii")
+
+    # load bitmap
+    bitmap = bytearray()
+    while True:
+        s = f.readline()
+        if not s or s[:7] == b"ENDCHAR":
+            break
+        bitmap += s[:-1]
+
+    # The word BBX
+    # followed by the width in x (BBw), height in y (BBh),
+    # and x and y displacement (BBxoff0, BByoff0)
+    # of the lower left corner from the origin of the character.
+    width, height, x_disp, y_disp = (int(p) for p in props["BBX"].split())
+
+    # The word DWIDTH
+    # followed by the width in x and y of the character in device pixels.
+    dwx, dwy = (int(p) for p in props["DWIDTH"].split())
+
+    bbox = (
+        (dwx, dwy),
+        (x_disp, -y_disp - height, width + x_disp, -y_disp),
+        (0, 0, width, height),
+    )
+
+    try:
+        im = Image.frombytes("1", (width, height), bitmap, "hex", "1")
+    except ValueError:
+        # deal with zero-width characters
+        im = Image.new("1", (width, height))
+
+    return id, int(props["ENCODING"]), bbox, im
+
+
+class BdfFontFile(FontFile.FontFile):
+    """Font file plugin for the X11 BDF format."""
+
+    def __init__(self, fp: BinaryIO) -> None:
+        super().__init__()
+
+        s = fp.readline()
+        if s[:13] != b"STARTFONT 2.1":
+            msg = "not a valid BDF file"
+            raise SyntaxError(msg)
+
+        props = {}
+        comments = []
+
+        while True:
+            s = fp.readline()
+            if not s or s[:13] == b"ENDPROPERTIES":
+                break
+            i = s.find(b" ")
+            props[s[:i].decode("ascii")] = s[i + 1 : -1].decode("ascii")
+            if s[:i] in [b"COMMENT", b"COPYRIGHT"]:
+                if s.find(b"LogicalFontDescription") < 0:
+                    comments.append(s[i + 1 : -1].decode("ascii"))
+
+        while True:
+            c = bdf_char(fp)
+            if not c:
+                break
+            id, ch, (xy, dst, src), im = c
+            if 0 <= ch < len(self.glyph):
+                self.glyph[ch] = xy, dst, src, im

venv/Lib/site-packages/PIL/BlpImagePlugin.py

@@ -0,0 +1,501 @@
+"""
+Blizzard Mipmap Format (.blp)
+Jerome Leclanche <jerome@leclan.ch>
+
+The contents of this file are hereby released in the public domain (CC0)
+Full text of the CC0 license:
+  https://creativecommons.org/publicdomain/zero/1.0/
+
+BLP1 files, used mostly in Warcraft III, are not fully supported.
+All types of BLP2 files used in World of Warcraft are supported.
+
+The BLP file structure consists of a header, up to 16 mipmaps of the
+texture
+
+Texture sizes must be powers of two, though the two dimensions do
+not have to be equal; 512x256 is valid, but 512x200 is not.
+The first mipmap (mipmap #0) is the full size image; each subsequent
+mipmap halves both dimensions. The final mipmap should be 1x1.
+
+BLP files come in many different flavours:
+* JPEG-compressed (type == 0) - only supported for BLP1.
+* RAW images (type == 1, encoding == 1). Each mipmap is stored as an
+  array of 8-bit values, one per pixel, left to right, top to bottom.
+  Each value is an index to the palette.
+* DXT-compressed (type == 1, encoding == 2):
+- DXT1 compression is used if alpha_encoding == 0.
+  - An additional alpha bit is used if alpha_depth == 1.
+  - DXT3 compression is used if alpha_encoding == 1.
+  - DXT5 compression is used if alpha_encoding == 7.
+"""
+
+from __future__ import annotations
+
+import abc
+import os
+import struct
+from enum import IntEnum
+from io import BytesIO
+from typing import IO
+
+from . import Image, ImageFile
+
+
+class Format(IntEnum):
+    JPEG = 0
+
+
+class Encoding(IntEnum):
+    UNCOMPRESSED = 1
+    DXT = 2
+    UNCOMPRESSED_RAW_BGRA = 3
+
+
+class AlphaEncoding(IntEnum):
+    DXT1 = 0
+    DXT3 = 1
+    DXT5 = 7
+
+
+def unpack_565(i: int) -> tuple[int, int, int]:
+    return ((i >> 11) & 0x1F) << 3, ((i >> 5) & 0x3F) << 2, (i & 0x1F) << 3
+
+
+def decode_dxt1(
+    data: bytes, alpha: bool = False
+) -> tuple[bytearray, bytearray, bytearray, bytearray]:
+    """
+    input: one "row" of data (i.e. will produce 4*width pixels)
+    """
+
+    blocks = len(data) // 8  # number of blocks in row
+    ret = (bytearray(), bytearray(), bytearray(), bytearray())
+
+    for block_index in range(blocks):
+        # Decode next 8-byte block.
+        idx = block_index * 8
+        color0, color1, bits = struct.unpack_from("<HHI", data, idx)
+
+        r0, g0, b0 = unpack_565(color0)
+        r1, g1, b1 = unpack_565(color1)
+
+        # Decode this block into 4x4 pixels
+        # Accumulate the results onto our 4 row accumulators
+        for j in range(4):
+            for i in range(4):
+                # get next control op and generate a pixel
+
+                control = bits & 3
+                bits = bits >> 2
+
+                a = 0xFF
+                if control == 0:
+                    r, g, b = r0, g0, b0
+                elif control == 1:
+                    r, g, b = r1, g1, b1
+                elif control == 2:
+                    if color0 > color1:
+                        r = (2 * r0 + r1) // 3
+                        g = (2 * g0 + g1) // 3
+                        b = (2 * b0 + b1) // 3
+                    else:
+                        r = (r0 + r1) // 2
+                        g = (g0 + g1) // 2
+                        b = (b0 + b1) // 2
+                elif control == 3:
+                    if color0 > color1:
+                        r = (2 * r1 + r0) // 3
+                        g = (2 * g1 + g0) // 3
+                        b = (2 * b1 + b0) // 3
+                    else:
+                        r, g, b, a = 0, 0, 0, 0
+
+                if alpha:
+                    ret[j].extend([r, g, b, a])
+                else:
+                    ret[j].extend([r, g, b])
+
+    return ret
+
+
+def decode_dxt3(data: bytes) -> tuple[bytearray, bytearray, bytearray, bytearray]:
+    """
+    input: one "row" of data (i.e. will produce 4*width pixels)
+    """
+
+    blocks = len(data) // 16  # number of blocks in row
+    ret = (bytearray(), bytearray(), bytearray(), bytearray())
+
+    for block_index in range(blocks):
+        idx = block_index * 16
+        block = data[idx : idx + 16]
+        # Decode next 16-byte block.
+        bits = struct.unpack_from("<8B", block)
+        color0, color1 = struct.unpack_from("<HH", block, 8)
+
+        (code,) = struct.unpack_from("<I", block, 12)
+
+        r0, g0, b0 = unpack_565(color0)
+        r1, g1, b1 = unpack_565(color1)
+
+        for j in range(4):
+            high = False  # Do we want the higher bits?
+            for i in range(4):
+                alphacode_index = (4 * j + i) // 2
+                a = bits[alphacode_index]
+                if high:
+                    high = False
+                    a >>= 4
+                else:
+                    high = True
+                    a &= 0xF
+                a *= 17  # We get a value between 0 and 15
+
+                color_code = (code >> 2 * (4 * j + i)) & 0x03
+
+                if color_code == 0:
+                    r, g, b = r0, g0, b0
+                elif color_code == 1:
+                    r, g, b = r1, g1, b1
+                elif color_code == 2:
+                    r = (2 * r0 + r1) // 3
+                    g = (2 * g0 + g1) // 3
+                    b = (2 * b0 + b1) // 3
+                elif color_code == 3:
+                    r = (2 * r1 + r0) // 3
+                    g = (2 * g1 + g0) // 3
+                    b = (2 * b1 + b0) // 3
+
+                ret[j].extend([r, g, b, a])
+
+    return ret
+
+
+def decode_dxt5(data: bytes) -> tuple[bytearray, bytearray, bytearray, bytearray]:
+    """
+    input: one "row" of data (i.e. will produce 4 * width pixels)
+    """
+
+    blocks = len(data) // 16  # number of blocks in row
+    ret = (bytearray(), bytearray(), bytearray(), bytearray())
+
+    for block_index in range(blocks):
+        idx = block_index * 16
+        block = data[idx : idx + 16]
+        # Decode next 16-byte block.
+        a0, a1 = struct.unpack_from("<BB", block)
+
+        bits = struct.unpack_from("<6B", block, 2)
+        alphacode1 = bits[2] | (bits[3] << 8) | (bits[4] << 16) | (bits[5] << 24)
+        alphacode2 = bits[0] | (bits[1] << 8)
+
+        color0, color1 = struct.unpack_from("<HH", block, 8)
+
+        (code,) = struct.unpack_from("<I", block, 12)
+
+        r0, g0, b0 = unpack_565(color0)
+        r1, g1, b1 = unpack_565(color1)
+
+        for j in range(4):
+            for i in range(4):
+                # get next control op and generate a pixel
+                alphacode_index = 3 * (4 * j + i)
+
+                if alphacode_index <= 12:
+                    alphacode = (alphacode2 >> alphacode_index) & 0x07
+                elif alphacode_index == 15:
+                    alphacode = (alphacode2 >> 15) | ((alphacode1 << 1) & 0x06)
+                else:  # alphacode_index >= 18 and alphacode_index <= 45
+                    alphacode = (alphacode1 >> (alphacode_index - 16)) & 0x07
+
+                if alphacode == 0:
+                    a = a0
+                elif alphacode == 1:
+                    a = a1
+                elif a0 > a1:
+                    a = ((8 - alphacode) * a0 + (alphacode - 1) * a1) // 7
+                elif alphacode == 6:
+                    a = 0
+                elif alphacode == 7:
+                    a = 255
+                else:
+                    a = ((6 - alphacode) * a0 + (alphacode - 1) * a1) // 5
+
+                color_code = (code >> 2 * (4 * j + i)) & 0x03
+
+                if color_code == 0:
+                    r, g, b = r0, g0, b0
+                elif color_code == 1:
+                    r, g, b = r1, g1, b1
+                elif color_code == 2:
+                    r = (2 * r0 + r1) // 3
+                    g = (2 * g0 + g1) // 3
+                    b = (2 * b0 + b1) // 3
+                elif color_code == 3:
+                    r = (2 * r1 + r0) // 3
+                    g = (2 * g1 + g0) // 3
+                    b = (2 * b1 + b0) // 3
+
+                ret[j].extend([r, g, b, a])
+
+    return ret
+
+
+class BLPFormatError(NotImplementedError):
+    pass
+
+
+def _accept(prefix: bytes) -> bool:
+    return prefix[:4] in (b"BLP1", b"BLP2")
+
+
+class BlpImageFile(ImageFile.ImageFile):
+    """
+    Blizzard Mipmap Format
+    """
+
+    format = "BLP"
+    format_description = "Blizzard Mipmap Format"
+
+    def _open(self) -> None:
+        self.magic = self.fp.read(4)
+        if not _accept(self.magic):
+            msg = f"Bad BLP magic {repr(self.magic)}"
+            raise BLPFormatError(msg)
+
+        compression = struct.unpack("<i", self.fp.read(4))[0]
+        if self.magic == b"BLP1":
+            alpha = struct.unpack("<I", self.fp.read(4))[0] != 0
+        else:
+            encoding = struct.unpack("<b", self.fp.read(1))[0]
+            alpha = struct.unpack("<b", self.fp.read(1))[0] != 0
+            alpha_encoding = struct.unpack("<b", self.fp.read(1))[0]
+            self.fp.seek(1, os.SEEK_CUR)  # mips
+
+        self._size = struct.unpack("<II", self.fp.read(8))
+
+        args: tuple[int, int, bool] | tuple[int, int, bool, int]
+        if self.magic == b"BLP1":
+            encoding = struct.unpack("<i", self.fp.read(4))[0]
+            self.fp.seek(4, os.SEEK_CUR)  # subtype
+
+            args = (compression, encoding, alpha)
+            offset = 28
+        else:
+            args = (compression, encoding, alpha, alpha_encoding)
+            offset = 20
+
+        decoder = self.magic.decode()
+
+        self._mode = "RGBA" if alpha else "RGB"
+        self.tile = [ImageFile._Tile(decoder, (0, 0) + self.size, offset, args)]
+
+
+class _BLPBaseDecoder(ImageFile.PyDecoder):
+    _pulls_fd = True
+
+    def decode(self, buffer: bytes | Image.SupportsArrayInterface) -> tuple[int, int]:
+        try:
+            self._read_header()
+            self._load()
+        except struct.error as e:
+            msg = "Truncated BLP file"
+            raise OSError(msg) from e
+        return -1, 0
+
+    @abc.abstractmethod
+    def _load(self) -> None:
+        pass
+
+    def _read_header(self) -> None:
+        self._offsets = struct.unpack("<16I", self._safe_read(16 * 4))
+        self._lengths = struct.unpack("<16I", self._safe_read(16 * 4))
+
+    def _safe_read(self, length: int) -> bytes:
+        assert self.fd is not None
+        return ImageFile._safe_read(self.fd, length)
+
+    def _read_palette(self) -> list[tuple[int, int, int, int]]:
+        ret = []
+        for i in range(256):
+            try:
+                b, g, r, a = struct.unpack("<4B", self._safe_read(4))
+            except struct.error:
+                break
+            ret.append((b, g, r, a))
+        return ret
+
+    def _read_bgra(
+        self, palette: list[tuple[int, int, int, int]], alpha: bool
+    ) -> bytearray:
+        data = bytearray()
+        _data = BytesIO(self._safe_read(self._lengths[0]))
+        while True:
+            try:
+                (offset,) = struct.unpack("<B", _data.read(1))
+            except struct.error:
+                break
+            b, g, r, a = palette[offset]
+            d: tuple[int, ...] = (r, g, b)
+            if alpha:
+                d += (a,)
+            data.extend(d)
+        return data
+
+
+class BLP1Decoder(_BLPBaseDecoder):
+    def _load(self) -> None:
+        self._compression, self._encoding, alpha = self.args
+
+        if self._compression == Format.JPEG:
+            self._decode_jpeg_stream()
+
+        elif self._compression == 1:
+            if self._encoding in (4, 5):
+                palette = self._read_palette()
+                data = self._read_bgra(palette, alpha)
+                self.set_as_raw(data)
+            else:
+                msg = f"Unsupported BLP encoding {repr(self._encoding)}"
+                raise BLPFormatError(msg)
+        else:
+            msg = f"Unsupported BLP compression {repr(self._encoding)}"
+            raise BLPFormatError(msg)
+
+    def _decode_jpeg_stream(self) -> None:
+        from .JpegImagePlugin import JpegImageFile
+
+        (jpeg_header_size,) = struct.unpack("<I", self._safe_read(4))
+        jpeg_header = self._safe_read(jpeg_header_size)
+        assert self.fd is not None
+        self._safe_read(self._offsets[0] - self.fd.tell())  # What IS this?
+        data = self._safe_read(self._lengths[0])
+        data = jpeg_header + data
+        image = JpegImageFile(BytesIO(data))
+        Image._decompression_bomb_check(image.size)
+        if image.mode == "CMYK":
+            decoder_name, extents, offset, args = image.tile[0]
+            assert isinstance(args, tuple)
+            image.tile = [
+                ImageFile._Tile(decoder_name, extents, offset, (args[0], "CMYK"))
+            ]
+        r, g, b = image.convert("RGB").split()
+        reversed_image = Image.merge("RGB", (b, g, r))
+        self.set_as_raw(reversed_image.tobytes())
+
+
+class BLP2Decoder(_BLPBaseDecoder):
+    def _load(self) -> None:
+        self._compression, self._encoding, alpha, self._alpha_encoding = self.args
+
+        palette = self._read_palette()
+
+        assert self.fd is not None
+        self.fd.seek(self._offsets[0])
+
+        if self._compression == 1:
+            # Uncompressed or DirectX compression
+
+            if self._encoding == Encoding.UNCOMPRESSED:
+                data = self._read_bgra(palette, alpha)
+
+            elif self._encoding == Encoding.DXT:
+                data = bytearray()
+                if self._alpha_encoding == AlphaEncoding.DXT1:
+                    linesize = (self.state.xsize + 3) // 4 * 8
+                    for yb in range((self.state.ysize + 3) // 4):
+                        for d in decode_dxt1(self._safe_read(linesize), alpha):
+                            data += d
+
+                elif self._alpha_encoding == AlphaEncoding.DXT3:
+                    linesize = (self.state.xsize + 3) // 4 * 16
+                    for yb in range((self.state.ysize + 3) // 4):
+                        for d in decode_dxt3(self._safe_read(linesize)):
+                            data += d
+
+                elif self._alpha_encoding == AlphaEncoding.DXT5:
+                    linesize = (self.state.xsize + 3) // 4 * 16
+                    for yb in range((self.state.ysize + 3) // 4):
+                        for d in decode_dxt5(self._safe_read(linesize)):
+                            data += d
+                else:
+                    msg = f"Unsupported alpha encoding {repr(self._alpha_encoding)}"
+                    raise BLPFormatError(msg)
+            else:
+                msg = f"Unknown BLP encoding {repr(self._encoding)}"
+                raise BLPFormatError(msg)
+
+        else:
+            msg = f"Unknown BLP compression {repr(self._compression)}"
+            raise BLPFormatError(msg)
+
+        self.set_as_raw(data)
+
+
+class BLPEncoder(ImageFile.PyEncoder):
+    _pushes_fd = True
+
+    def _write_palette(self) -> bytes:
+        data = b""
+        assert self.im is not None
+        palette = self.im.getpalette("RGBA", "RGBA")
+        for i in range(len(palette) // 4):
+            r, g, b, a = palette[i * 4 : (i + 1) * 4]
+            data += struct.pack("<4B", b, g, r, a)
+        while len(data) < 256 * 4:
+            data += b"\x00" * 4
+        return data
+
+    def encode(self, bufsize: int) -> tuple[int, int, bytes]:
+        palette_data = self._write_palette()
+
+        offset = 20 + 16 * 4 * 2 + len(palette_data)
+        data = struct.pack("<16I", offset, *((0,) * 15))
+
+        assert self.im is not None
+        w, h = self.im.size
+        data += struct.pack("<16I", w * h, *((0,) * 15))
+
+        data += palette_data
+
+        for y in range(h):
+            for x in range(w):
+                data += struct.pack("<B", self.im.getpixel((x, y)))
+
+        return len(data), 0, data
+
+
+def _save(im: Image.Image, fp: IO[bytes], filename: str | bytes) -> None:
+    if im.mode != "P":
+        msg = "Unsupported BLP image mode"
+        raise ValueError(msg)
+
+    magic = b"BLP1" if im.encoderinfo.get("blp_version") == "BLP1" else b"BLP2"
+    fp.write(magic)
+
+    assert im.palette is not None
+    fp.write(struct.pack("<i", 1))  # Uncompressed or DirectX compression
+
+    alpha_depth = 1 if im.palette.mode == "RGBA" else 0
+    if magic == b"BLP1":
+        fp.write(struct.pack("<L", alpha_depth))
+    else:
+        fp.write(struct.pack("<b", Encoding.UNCOMPRESSED))
+        fp.write(struct.pack("<b", alpha_depth))
+        fp.write(struct.pack("<b", 0))  # alpha encoding
+        fp.write(struct.pack("<b", 0))  # mips
+    fp.write(struct.pack("<II", *im.size))
+    if magic == b"BLP1":
+        fp.write(struct.pack("<i", 5))
+        fp.write(struct.pack("<i", 0))
+
+    ImageFile._save(im, fp, [ImageFile._Tile("BLP", (0, 0) + im.size, 0, im.mode)])
+
+
+Image.register_open(BlpImageFile.format, BlpImageFile, _accept)
+Image.register_extension(BlpImageFile.format, ".blp")
+Image.register_decoder("BLP1", BLP1Decoder)
+Image.register_decoder("BLP2", BLP2Decoder)
+
+Image.register_save(BlpImageFile.format, _save)
+Image.register_encoder("BLP", BLPEncoder)

venv/Lib/site-packages/PIL/BmpImagePlugin.py

@@ -0,0 +1,511 @@
+#
+# The Python Imaging Library.
+# $Id$
+#
+# BMP file handler
+#
+# Windows (and OS/2) native bitmap storage format.
+#
+# history:
+# 1995-09-01 fl   Created
+# 1996-04-30 fl   Added save
+# 1997-08-27 fl   Fixed save of 1-bit images
+# 1998-03-06 fl   Load P images as L where possible
+# 1998-07-03 fl   Load P images as 1 where possible
+# 1998-12-29 fl   Handle small palettes
+# 2002-12-30 fl   Fixed load of 1-bit palette images
+# 2003-04-21 fl   Fixed load of 1-bit monochrome images
+# 2003-04-23 fl   Added limited support for BI_BITFIELDS compression
+#
+# Copyright (c) 1997-2003 by Secret Labs AB
+# Copyright (c) 1995-2003 by Fredrik Lundh
+#
+# See the README file for information on usage and redistribution.
+#
+from __future__ import annotations
+
+import os
+from typing import IO, Any
+
+from . import Image, ImageFile, ImagePalette
+from ._binary import i16le as i16
+from ._binary import i32le as i32
+from ._binary import o8
+from ._binary import o16le as o16
+from ._binary import o32le as o32
+
+#
+# --------------------------------------------------------------------
+# Read BMP file
+
+BIT2MODE = {
+    # bits => mode, rawmode
+    1: ("P", "P;1"),
+    4: ("P", "P;4"),
+    8: ("P", "P"),
+    16: ("RGB", "BGR;15"),
+    24: ("RGB", "BGR"),
+    32: ("RGB", "BGRX"),
+}
+
+
+def _accept(prefix: bytes) -> bool:
+    return prefix[:2] == b"BM"
+
+
+def _dib_accept(prefix: bytes) -> bool:
+    return i32(prefix) in [12, 40, 52, 56, 64, 108, 124]
+
+
+# =============================================================================
+# Image plugin for the Windows BMP format.
+# =============================================================================
+class BmpImageFile(ImageFile.ImageFile):
+    """Image plugin for the Windows Bitmap format (BMP)"""
+
+    # ------------------------------------------------------------- Description
+    format_description = "Windows Bitmap"
+    format = "BMP"
+
+    # -------------------------------------------------- BMP Compression values
+    COMPRESSIONS = {"RAW": 0, "RLE8": 1, "RLE4": 2, "BITFIELDS": 3, "JPEG": 4, "PNG": 5}
+    for k, v in COMPRESSIONS.items():
+        vars()[k] = v
+
+    def _bitmap(self, header: int = 0, offset: int = 0) -> None:
+        """Read relevant info about the BMP"""
+        read, seek = self.fp.read, self.fp.seek
+        if header:
+            seek(header)
+        # read bmp header size @offset 14 (this is part of the header size)
+        file_info: dict[str, bool | int | tuple[int, ...]] = {
+            "header_size": i32(read(4)),
+            "direction": -1,
+        }
+
+        # -------------------- If requested, read header at a specific position
+        # read the rest of the bmp header, without its size
+        assert isinstance(file_info["header_size"], int)
+        header_data = ImageFile._safe_read(self.fp, file_info["header_size"] - 4)
+
+        # ------------------------------- Windows Bitmap v2, IBM OS/2 Bitmap v1
+        # ----- This format has different offsets because of width/height types
+        # 12: BITMAPCOREHEADER/OS21XBITMAPHEADER
+        if file_info["header_size"] == 12:
+            file_info["width"] = i16(header_data, 0)
+            file_info["height"] = i16(header_data, 2)
+            file_info["planes"] = i16(header_data, 4)
+            file_info["bits"] = i16(header_data, 6)
+            file_info["compression"] = self.COMPRESSIONS["RAW"]
+            file_info["palette_padding"] = 3
+
+        # --------------------------------------------- Windows Bitmap v3 to v5
+        #  40: BITMAPINFOHEADER
+        #  52: BITMAPV2HEADER
+        #  56: BITMAPV3HEADER
+        #  64: BITMAPCOREHEADER2/OS22XBITMAPHEADER
+        # 108: BITMAPV4HEADER
+        # 124: BITMAPV5HEADER
+        elif file_info["header_size"] in (40, 52, 56, 64, 108, 124):
+            file_info["y_flip"] = header_data[7] == 0xFF
+            file_info["direction"] = 1 if file_info["y_flip"] else -1
+            file_info["width"] = i32(header_data, 0)
+            file_info["height"] = (
+                i32(header_data, 4)
+                if not file_info["y_flip"]
+                else 2**32 - i32(header_data, 4)
+            )
+            file_info["planes"] = i16(header_data, 8)
+            file_info["bits"] = i16(header_data, 10)
+            file_info["compression"] = i32(header_data, 12)
+            # byte size of pixel data
+            file_info["data_size"] = i32(header_data, 16)
+            file_info["pixels_per_meter"] = (
+                i32(header_data, 20),
+                i32(header_data, 24),
+            )
+            file_info["colors"] = i32(header_data, 28)
+            file_info["palette_padding"] = 4
+            assert isinstance(file_info["pixels_per_meter"], tuple)
+            self.info["dpi"] = tuple(x / 39.3701 for x in file_info["pixels_per_meter"])
+            if file_info["compression"] == self.COMPRESSIONS["BITFIELDS"]:
+                masks = ["r_mask", "g_mask", "b_mask"]
+                if len(header_data) >= 48:
+                    if len(header_data) >= 52:
+                        masks.append("a_mask")
+                    else:
+                        file_info["a_mask"] = 0x0
+                    for idx, mask in enumerate(masks):
+                        file_info[mask] = i32(header_data, 36 + idx * 4)
+                else:
+                    # 40 byte headers only have the three components in the
+                    # bitfields masks, ref:
+                    # https://msdn.microsoft.com/en-us/library/windows/desktop/dd183376(v=vs.85).aspx
+                    # See also
+                    # https://github.com/python-pillow/Pillow/issues/1293
+                    # There is a 4th component in the RGBQuad, in the alpha
+                    # location, but it is listed as a reserved component,
+                    # and it is not generally an alpha channel
+                    file_info["a_mask"] = 0x0
+                    for mask in masks:
+                        file_info[mask] = i32(read(4))
+                assert isinstance(file_info["r_mask"], int)
+                assert isinstance(file_info["g_mask"], int)
+                assert isinstance(file_info["b_mask"], int)
+                assert isinstance(file_info["a_mask"], int)
+                file_info["rgb_mask"] = (
+                    file_info["r_mask"],
+                    file_info["g_mask"],
+                    file_info["b_mask"],
+                )
+                file_info["rgba_mask"] = (
+                    file_info["r_mask"],
+                    file_info["g_mask"],
+                    file_info["b_mask"],
+                    file_info["a_mask"],
+                )
+        else:
+            msg = f"Unsupported BMP header type ({file_info['header_size']})"
+            raise OSError(msg)
+
+        # ------------------ Special case : header is reported 40, which
+        # ---------------------- is shorter than real size for bpp >= 16
+        assert isinstance(file_info["width"], int)
+        assert isinstance(file_info["height"], int)
+        self._size = file_info["width"], file_info["height"]
+
+        # ------- If color count was not found in the header, compute from bits
+        assert isinstance(file_info["bits"], int)
+        file_info["colors"] = (
+            file_info["colors"]
+            if file_info.get("colors", 0)
+            else (1 << file_info["bits"])
+        )
+        assert isinstance(file_info["colors"], int)
+        if offset == 14 + file_info["header_size"] and file_info["bits"] <= 8:
+            offset += 4 * file_info["colors"]
+
+        # ---------------------- Check bit depth for unusual unsupported values
+        self._mode, raw_mode = BIT2MODE.get(file_info["bits"], ("", ""))
+        if not self.mode:
+            msg = f"Unsupported BMP pixel depth ({file_info['bits']})"
+            raise OSError(msg)
+
+        # ---------------- Process BMP with Bitfields compression (not palette)
+        decoder_name = "raw"
+        if file_info["compression"] == self.COMPRESSIONS["BITFIELDS"]:
+            SUPPORTED: dict[int, list[tuple[int, ...]]] = {
+                32: [
+                    (0xFF0000, 0xFF00, 0xFF, 0x0),
+                    (0xFF000000, 0xFF0000, 0xFF00, 0x0),
+                    (0xFF000000, 0xFF00, 0xFF, 0x0),
+                    (0xFF000000, 0xFF0000, 0xFF00, 0xFF),
+                    (0xFF, 0xFF00, 0xFF0000, 0xFF000000),
+                    (0xFF0000, 0xFF00, 0xFF, 0xFF000000),
+                    (0xFF000000, 0xFF00, 0xFF, 0xFF0000),
+                    (0x0, 0x0, 0x0, 0x0),
+                ],
+                24: [(0xFF0000, 0xFF00, 0xFF)],
+                16: [(0xF800, 0x7E0, 0x1F), (0x7C00, 0x3E0, 0x1F)],
+            }
+            MASK_MODES = {
+                (32, (0xFF0000, 0xFF00, 0xFF, 0x0)): "BGRX",
+                (32, (0xFF000000, 0xFF0000, 0xFF00, 0x0)): "XBGR",
+                (32, (0xFF000000, 0xFF00, 0xFF, 0x0)): "BGXR",
+                (32, (0xFF000000, 0xFF0000, 0xFF00, 0xFF)): "ABGR",
+                (32, (0xFF, 0xFF00, 0xFF0000, 0xFF000000)): "RGBA",
+                (32, (0xFF0000, 0xFF00, 0xFF, 0xFF000000)): "BGRA",
+                (32, (0xFF000000, 0xFF00, 0xFF, 0xFF0000)): "BGAR",
+                (32, (0x0, 0x0, 0x0, 0x0)): "BGRA",
+                (24, (0xFF0000, 0xFF00, 0xFF)): "BGR",
+                (16, (0xF800, 0x7E0, 0x1F)): "BGR;16",
+                (16, (0x7C00, 0x3E0, 0x1F)): "BGR;15",
+            }
+            if file_info["bits"] in SUPPORTED:
+                if (
+                    file_info["bits"] == 32
+                    and file_info["rgba_mask"] in SUPPORTED[file_info["bits"]]
+                ):
+                    assert isinstance(file_info["rgba_mask"], tuple)
+                    raw_mode = MASK_MODES[(file_info["bits"], file_info["rgba_mask"])]
+                    self._mode = "RGBA" if "A" in raw_mode else self.mode
+                elif (
+                    file_info["bits"] in (24, 16)
+                    and file_info["rgb_mask"] in SUPPORTED[file_info["bits"]]
+                ):
+                    assert isinstance(file_info["rgb_mask"], tuple)
+                    raw_mode = MASK_MODES[(file_info["bits"], file_info["rgb_mask"])]
+                else:
+                    msg = "Unsupported BMP bitfields layout"
+                    raise OSError(msg)
+            else:
+                msg = "Unsupported BMP bitfields layout"
+                raise OSError(msg)
+        elif file_info["compression"] == self.COMPRESSIONS["RAW"]:
+            if file_info["bits"] == 32 and header == 22:  # 32-bit .cur offset
+                raw_mode, self._mode = "BGRA", "RGBA"
+        elif file_info["compression"] in (
+            self.COMPRESSIONS["RLE8"],
+            self.COMPRESSIONS["RLE4"],
+        ):
+            decoder_name = "bmp_rle"
+        else:
+            msg = f"Unsupported BMP compression ({file_info['compression']})"
+            raise OSError(msg)
+
+        # --------------- Once the header is processed, process the palette/LUT
+        if self.mode == "P":  # Paletted for 1, 4 and 8 bit images
+            # ---------------------------------------------------- 1-bit images
+            if not (0 < file_info["colors"] <= 65536):
+                msg = f"Unsupported BMP Palette size ({file_info['colors']})"
+                raise OSError(msg)
+            else:
+                assert isinstance(file_info["palette_padding"], int)
+                padding = file_info["palette_padding"]
+                palette = read(padding * file_info["colors"])
+                grayscale = True
+                indices = (
+                    (0, 255)
+                    if file_info["colors"] == 2
+                    else list(range(file_info["colors"]))
+                )
+
+                # ----------------- Check if grayscale and ignore palette if so
+                for ind, val in enumerate(indices):
+                    rgb = palette[ind * padding : ind * padding + 3]
+                    if rgb != o8(val) * 3:
+                        grayscale = False
+
+                # ------- If all colors are gray, white or black, ditch palette
+                if grayscale:
+                    self._mode = "1" if file_info["colors"] == 2 else "L"
+                    raw_mode = self.mode
+                else:
+                    self._mode = "P"
+                    self.palette = ImagePalette.raw(
+                        "BGRX" if padding == 4 else "BGR", palette
+                    )
+
+        # ---------------------------- Finally set the tile data for the plugin
+        self.info["compression"] = file_info["compression"]
+        args: list[Any] = [raw_mode]
+        if decoder_name == "bmp_rle":
+            args.append(file_info["compression"] == self.COMPRESSIONS["RLE4"])
+        else:
+            assert isinstance(file_info["width"], int)
+            args.append(((file_info["width"] * file_info["bits"] + 31) >> 3) & (~3))
+        args.append(file_info["direction"])
+        self.tile = [
+            ImageFile._Tile(
+                decoder_name,
+                (0, 0, file_info["width"], file_info["height"]),
+                offset or self.fp.tell(),
+                tuple(args),
+            )
+        ]
+
+    def _open(self) -> None:
+        """Open file, check magic number and read header"""
+        # read 14 bytes: magic number, filesize, reserved, header final offset
+        head_data = self.fp.read(14)
+        # choke if the file does not have the required magic bytes
+        if not _accept(head_data):
+            msg = "Not a BMP file"
+            raise SyntaxError(msg)
+        # read the start position of the BMP image data (u32)
+        offset = i32(head_data, 10)
+        # load bitmap information (offset=raster info)
+        self._bitmap(offset=offset)
+
+
+class BmpRleDecoder(ImageFile.PyDecoder):
+    _pulls_fd = True
+
+    def decode(self, buffer: bytes | Image.SupportsArrayInterface) -> tuple[int, int]:
+        assert self.fd is not None
+        rle4 = self.args[1]
+        data = bytearray()
+        x = 0
+        dest_length = self.state.xsize * self.state.ysize
+        while len(data) < dest_length:
+            pixels = self.fd.read(1)
+            byte = self.fd.read(1)
+            if not pixels or not byte:
+                break
+            num_pixels = pixels[0]
+            if num_pixels:
+                # encoded mode
+                if x + num_pixels > self.state.xsize:
+                    # Too much data for row
+                    num_pixels = max(0, self.state.xsize - x)
+                if rle4:
+                    first_pixel = o8(byte[0] >> 4)
+                    second_pixel = o8(byte[0] & 0x0F)
+                    for index in range(num_pixels):
+                        if index % 2 == 0:
+                            data += first_pixel
+                        else:
+                            data += second_pixel
+                else:
+                    data += byte * num_pixels
+                x += num_pixels
+            else:
+                if byte[0] == 0:
+                    # end of line
+                    while len(data) % self.state.xsize != 0:
+                        data += b"\x00"
+                    x = 0
+                elif byte[0] == 1:
+                    # end of bitmap
+                    break
+                elif byte[0] == 2:
+                    # delta
+                    bytes_read = self.fd.read(2)
+                    if len(bytes_read) < 2:
+                        break
+                    right, up = self.fd.read(2)
+                    data += b"\x00" * (right + up * self.state.xsize)
+                    x = len(data) % self.state.xsize
+                else:
+                    # absolute mode
+                    if rle4:
+                        # 2 pixels per byte
+                        byte_count = byte[0] // 2
+                        bytes_read = self.fd.read(byte_count)
+                        for byte_read in bytes_read:
+                            data += o8(byte_read >> 4)
+                            data += o8(byte_read & 0x0F)
+                    else:
+                        byte_count = byte[0]
+                        bytes_read = self.fd.read(byte_count)
+                        data += bytes_read
+                    if len(bytes_read) < byte_count:
+                        break
+                    x += byte[0]
+
+                    # align to 16-bit word boundary
+                    if self.fd.tell() % 2 != 0:
+                        self.fd.seek(1, os.SEEK_CUR)
+        rawmode = "L" if self.mode == "L" else "P"
+        self.set_as_raw(bytes(data), rawmode, (0, self.args[-1]))
+        return -1, 0
+
+
+# =============================================================================
+# Image plugin for the DIB format (BMP alias)
+# =============================================================================
+class DibImageFile(BmpImageFile):
+    format = "DIB"
+    format_description = "Windows Bitmap"
+
+    def _open(self) -> None:
+        self._bitmap()
+
+
+#
+# --------------------------------------------------------------------
+# Write BMP file
+
+
+SAVE = {
+    "1": ("1", 1, 2),
+    "L": ("L", 8, 256),
+    "P": ("P", 8, 256),
+    "RGB": ("BGR", 24, 0),
+    "RGBA": ("BGRA", 32, 0),
+}
+
+
+def _dib_save(im: Image.Image, fp: IO[bytes], filename: str | bytes) -> None:
+    _save(im, fp, filename, False)
+
+
+def _save(
+    im: Image.Image, fp: IO[bytes], filename: str | bytes, bitmap_header: bool = True
+) -> None:
+    try:
+        rawmode, bits, colors = SAVE[im.mode]
+    except KeyError as e:
+        msg = f"cannot write mode {im.mode} as BMP"
+        raise OSError(msg) from e
+
+    info = im.encoderinfo
+
+    dpi = info.get("dpi", (96, 96))
+
+    # 1 meter == 39.3701 inches
+    ppm = tuple(int(x * 39.3701 + 0.5) for x in dpi)
+
+    stride = ((im.size[0] * bits + 7) // 8 + 3) & (~3)
+    header = 40  # or 64 for OS/2 version 2
+    image = stride * im.size[1]
+
+    if im.mode == "1":
+        palette = b"".join(o8(i) * 4 for i in (0, 255))
+    elif im.mode == "L":
+        palette = b"".join(o8(i) * 4 for i in range(256))
+    elif im.mode == "P":
+        palette = im.im.getpalette("RGB", "BGRX")
+        colors = len(palette) // 4
+    else:
+        palette = None
+
+    # bitmap header
+    if bitmap_header:
+        offset = 14 + header + colors * 4
+        file_size = offset + image
+        if file_size > 2**32 - 1:
+            msg = "File size is too large for the BMP format"
+            raise ValueError(msg)
+        fp.write(
+            b"BM"  # file type (magic)
+            + o32(file_size)  # file size
+            + o32(0)  # reserved
+            + o32(offset)  # image data offset
+        )
+
+    # bitmap info header
+    fp.write(
+        o32(header)  # info header size
+        + o32(im.size[0])  # width
+        + o32(im.size[1])  # height
+        + o16(1)  # planes
+        + o16(bits)  # depth
+        + o32(0)  # compression (0=uncompressed)
+        + o32(image)  # size of bitmap
+        + o32(ppm[0])  # resolution
+        + o32(ppm[1])  # resolution
+        + o32(colors)  # colors used
+        + o32(colors)  # colors important
+    )
+
+    fp.write(b"\0" * (header - 40))  # padding (for OS/2 format)
+
+    if palette:
+        fp.write(palette)
+
+    ImageFile._save(
+        im, fp, [ImageFile._Tile("raw", (0, 0) + im.size, 0, (rawmode, stride, -1))]
+    )
+
+
+#
+# --------------------------------------------------------------------
+# Registry
+
+
+Image.register_open(BmpImageFile.format, BmpImageFile, _accept)
+Image.register_save(BmpImageFile.format, _save)
+
+Image.register_extension(BmpImageFile.format, ".bmp")
+
+Image.register_mime(BmpImageFile.format, "image/bmp")
+
+Image.register_decoder("bmp_rle", BmpRleDecoder)
+
+Image.register_open(DibImageFile.format, DibImageFile, _dib_accept)
+Image.register_save(DibImageFile.format, _dib_save)
+
+Image.register_extension(DibImageFile.format, ".dib")
+
+Image.register_mime(DibImageFile.format, "image/bmp")

venv/Lib/site-packages/PIL/BufrStubImagePlugin.py

@@ -0,0 +1,76 @@
+#
+# The Python Imaging Library
+# $Id$
+#
+# BUFR stub adapter
+#
+# Copyright (c) 1996-2003 by Fredrik Lundh
+#
+# See the README file for information on usage and redistribution.
+#
+from __future__ import annotations
+
+from typing import IO
+
+from . import Image, ImageFile
+
+_handler = None
+
+
+def register_handler(handler: ImageFile.StubHandler | None) -> None:
+    """
+    Install application-specific BUFR image handler.
+
+    :param handler: Handler object.
+    """
+    global _handler
+    _handler = handler
+
+
+# --------------------------------------------------------------------
+# Image adapter
+
+
+def _accept(prefix: bytes) -> bool:
+    return prefix[:4] == b"BUFR" or prefix[:4] == b"ZCZC"
+
+
+class BufrStubImageFile(ImageFile.StubImageFile):
+    format = "BUFR"
+    format_description = "BUFR"
+
+    def _open(self) -> None:
+        offset = self.fp.tell()
+
+        if not _accept(self.fp.read(4)):
+            msg = "Not a BUFR file"
+            raise SyntaxError(msg)
+
+        self.fp.seek(offset)
+
+        # make something up
+        self._mode = "F"
+        self._size = 1, 1
+
+        loader = self._load()
+        if loader:
+            loader.open(self)
+
+    def _load(self) -> ImageFile.StubHandler | None:
+        return _handler
+
+
+def _save(im: Image.Image, fp: IO[bytes], filename: str | bytes) -> None:
+    if _handler is None or not hasattr(_handler, "save"):
+        msg = "BUFR save handler not installed"
+        raise OSError(msg)
+    _handler.save(im, fp, filename)
+
+
+# --------------------------------------------------------------------
+# Registry
+
+Image.register_open(BufrStubImageFile.format, BufrStubImageFile, _accept)
+Image.register_save(BufrStubImageFile.format, _save)
+
+Image.register_extension(BufrStubImageFile.format, ".bufr")

venv/Lib/site-packages/PIL/ContainerIO.py

@@ -0,0 +1,173 @@
+#
+# The Python Imaging Library.
+# $Id$
+#
+# a class to read from a container file
+#
+# History:
+# 1995-06-18 fl     Created
+# 1995-09-07 fl     Added readline(), readlines()
+#
+# Copyright (c) 1997-2001 by Secret Labs AB
+# Copyright (c) 1995 by Fredrik Lundh
+#
+# See the README file for information on usage and redistribution.
+#
+from __future__ import annotations
+
+import io
+from collections.abc import Iterable
+from typing import IO, AnyStr, NoReturn
+
+
+class ContainerIO(IO[AnyStr]):
+    """
+    A file object that provides read access to a part of an existing
+    file (for example a TAR file).
+    """
+
+    def __init__(self, file: IO[AnyStr], offset: int, length: int) -> None:
+        """
+        Create file object.
+
+        :param file: Existing file.
+        :param offset: Start of region, in bytes.
+        :param length: Size of region, in bytes.
+        """
+        self.fh: IO[AnyStr] = file
+        self.pos = 0
+        self.offset = offset
+        self.length = length
+        self.fh.seek(offset)
+
+    ##
+    # Always false.
+
+    def isatty(self) -> bool:
+        return False
+
+    def seekable(self) -> bool:
+        return True
+
+    def seek(self, offset: int, mode: int = io.SEEK_SET) -> int:
+        """
+        Move file pointer.
+
+        :param offset: Offset in bytes.
+        :param mode: Starting position. Use 0 for beginning of region, 1
+           for current offset, and 2 for end of region.  You cannot move
+           the pointer outside the defined region.
+        :returns: Offset from start of region, in bytes.
+        """
+        if mode == 1:
+            self.pos = self.pos + offset
+        elif mode == 2:
+            self.pos = self.length + offset
+        else:
+            self.pos = offset
+        # clamp
+        self.pos = max(0, min(self.pos, self.length))
+        self.fh.seek(self.offset + self.pos)
+        return self.pos
+
+    def tell(self) -> int:
+        """
+        Get current file pointer.
+
+        :returns: Offset from start of region, in bytes.
+        """
+        return self.pos
+
+    def readable(self) -> bool:
+        return True
+
+    def read(self, n: int = -1) -> AnyStr:
+        """
+        Read data.
+
+        :param n: Number of bytes to read. If omitted, zero or negative,
+            read until end of region.
+        :returns: An 8-bit string.
+        """
+        if n > 0:
+            n = min(n, self.length - self.pos)
+        else:
+            n = self.length - self.pos
+        if n <= 0:  # EOF
+            return b"" if "b" in self.fh.mode else ""  # type: ignore[return-value]
+        self.pos = self.pos + n
+        return self.fh.read(n)
+
+    def readline(self, n: int = -1) -> AnyStr:
+        """
+        Read a line of text.
+
+        :param n: Number of bytes to read. If omitted, zero or negative,
+            read until end of line.
+        :returns: An 8-bit string.
+        """
+        s: AnyStr = b"" if "b" in self.fh.mode else ""  # type: ignore[assignment]
+        newline_character = b"\n" if "b" in self.fh.mode else "\n"
+        while True:
+            c = self.read(1)
+            if not c:
+                break
+            s = s + c
+            if c == newline_character or len(s) == n:
+                break
+        return s
+
+    def readlines(self, n: int | None = -1) -> list[AnyStr]:
+        """
+        Read multiple lines of text.
+
+        :param n: Number of lines to read. If omitted, zero, negative or None,
+            read until end of region.
+        :returns: A list of 8-bit strings.
+        """
+        lines = []
+        while True:
+            s = self.readline()
+            if not s:
+                break
+            lines.append(s)
+            if len(lines) == n:
+                break
+        return lines
+
+    def writable(self) -> bool:
+        return False
+
+    def write(self, b: AnyStr) -> NoReturn:
+        raise NotImplementedError()
+
+    def writelines(self, lines: Iterable[AnyStr]) -> NoReturn:
+        raise NotImplementedError()
+
+    def truncate(self, size: int | None = None) -> int:
+        raise NotImplementedError()
+
+    def __enter__(self) -> ContainerIO[AnyStr]:
+        return self
+
+    def __exit__(self, *args: object) -> None:
+        self.close()
+
+    def __iter__(self) -> ContainerIO[AnyStr]:
+        return self
+
+    def __next__(self) -> AnyStr:
+        line = self.readline()
+        if not line:
+            msg = "end of region"
+            raise StopIteration(msg)
+        return line
+
+    def fileno(self) -> int:
+        return self.fh.fileno()
+
+    def flush(self) -> None:
+        self.fh.flush()
+
+    def close(self) -> None:
+        self.fh.close()

venv/Lib/site-packages/PIL/CurImagePlugin.py

@@ -0,0 +1,75 @@
+#
+# The Python Imaging Library.
+# $Id$
+#
+# Windows Cursor support for PIL
+#
+# notes:
+#       uses BmpImagePlugin.py to read the bitmap data.
+#
+# history:
+#       96-05-27 fl     Created
+#
+# Copyright (c) Secret Labs AB 1997.
+# Copyright (c) Fredrik Lundh 1996.
+#
+# See the README file for information on usage and redistribution.
+#
+from __future__ import annotations
+
+from . import BmpImagePlugin, Image, ImageFile
+from ._binary import i16le as i16
+from ._binary import i32le as i32
+
+#
+# --------------------------------------------------------------------
+
+
+def _accept(prefix: bytes) -> bool:
+    return prefix[:4] == b"\0\0\2\0"
+
+
+##
+# Image plugin for Windows Cursor files.
+
+
+class CurImageFile(BmpImagePlugin.BmpImageFile):
+    format = "CUR"
+    format_description = "Windows Cursor"
+
+    def _open(self) -> None:
+        offset = self.fp.tell()
+
+        # check magic
+        s = self.fp.read(6)
+        if not _accept(s):
+            msg = "not a CUR file"
+            raise SyntaxError(msg)
+
+        # pick the largest cursor in the file
+        m = b""
+        for i in range(i16(s, 4)):
+            s = self.fp.read(16)
+            if not m:
+                m = s
+            elif s[0] > m[0] and s[1] > m[1]:
+                m = s
+        if not m:
+            msg = "No cursors were found"
+            raise TypeError(msg)
+
+        # load as bitmap
+        self._bitmap(i32(m, 12) + offset)
+
+        # patch up the bitmap height
+        self._size = self.size[0], self.size[1] // 2
+        d, e, o, a = self.tile[0]
+        self.tile[0] = ImageFile._Tile(d, (0, 0) + self.size, o, a)
+
+
+#
+# --------------------------------------------------------------------
+
+Image.register_open(CurImageFile.format, CurImageFile, _accept)
+
+Image.register_extension(CurImageFile.format, ".cur")

venv/Lib/site-packages/PIL/DcxImagePlugin.py

@@ -0,0 +1,80 @@
+#
+# The Python Imaging Library.
+# $Id$
+#
+# DCX file handling
+#
+# DCX is a container file format defined by Intel, commonly used
+# for fax applications.  Each DCX file consists of a directory
+# (a list of file offsets) followed by a set of (usually 1-bit)
+# PCX files.
+#
+# History:
+# 1995-09-09 fl   Created
+# 1996-03-20 fl   Properly derived from PcxImageFile.
+# 1998-07-15 fl   Renamed offset attribute to avoid name clash
+# 2002-07-30 fl   Fixed file handling
+#
+# Copyright (c) 1997-98 by Secret Labs AB.
+# Copyright (c) 1995-96 by Fredrik Lundh.
+#
+# See the README file for information on usage and redistribution.
+#
+from __future__ import annotations
+
+from . import Image
+from ._binary import i32le as i32
+from .PcxImagePlugin import PcxImageFile
+
+MAGIC = 0x3ADE68B1  # QUIZ: what's this value, then?
+
+
+def _accept(prefix: bytes) -> bool:
+    return len(prefix) >= 4 and i32(prefix) == MAGIC
+
+
+##
+# Image plugin for the Intel DCX format.
+
+
+class DcxImageFile(PcxImageFile):
+    format = "DCX"
+    format_description = "Intel DCX"
+    _close_exclusive_fp_after_loading = False
+
+    def _open(self) -> None:
+        # Header
+        s = self.fp.read(4)
+        if not _accept(s):
+            msg = "not a DCX file"
+            raise SyntaxError(msg)
+
+        # Component directory
+        self._offset = []
+        for i in range(1024):
+            offset = i32(self.fp.read(4))
+            if not offset:
+                break
+            self._offset.append(offset)
+
+        self._fp = self.fp
+        self.frame = -1
+        self.n_frames = len(self._offset)
+        self.is_animated = self.n_frames > 1
+        self.seek(0)
+
+    def seek(self, frame: int) -> None:
+        if not self._seek_check(frame):
+            return
+        self.frame = frame
+        self.fp = self._fp
+        self.fp.seek(self._offset[frame])
+        PcxImageFile._open(self)
+
+    def tell(self) -> int:
+        return self.frame
+
+
+Image.register_open(DcxImageFile.format, DcxImageFile, _accept)
+
+Image.register_extension(DcxImageFile.format, ".dcx")

venv/Lib/site-packages/PIL/DdsImagePlugin.py

@@ -0,0 +1,573 @@
+"""
+A Pillow loader for .dds files (S3TC-compressed aka DXTC)
+Jerome Leclanche <jerome@leclan.ch>
+
+Documentation:
+https://web.archive.org/web/20170802060935/http://oss.sgi.com/projects/ogl-sample/registry/EXT/texture_compression_s3tc.txt
+
+The contents of this file are hereby released in the public domain (CC0)
+Full text of the CC0 license:
+https://creativecommons.org/publicdomain/zero/1.0/
+"""
+
+from __future__ import annotations
+
+import io
+import struct
+import sys
+from enum import IntEnum, IntFlag
+from typing import IO
+
+from . import Image, ImageFile, ImagePalette
+from ._binary import i32le as i32
+from ._binary import o8
+from ._binary import o32le as o32
+
+# Magic ("DDS ")
+DDS_MAGIC = 0x20534444
+
+
+# DDS flags
+class DDSD(IntFlag):
+    CAPS = 0x1
+    HEIGHT = 0x2
+    WIDTH = 0x4
+    PITCH = 0x8
+    PIXELFORMAT = 0x1000
+    MIPMAPCOUNT = 0x20000
+    LINEARSIZE = 0x80000
+    DEPTH = 0x800000
+
+
+# DDS caps
+class DDSCAPS(IntFlag):
+    COMPLEX = 0x8
+    TEXTURE = 0x1000
+    MIPMAP = 0x400000
+
+
+class DDSCAPS2(IntFlag):
+    CUBEMAP = 0x200
+    CUBEMAP_POSITIVEX = 0x400
+    CUBEMAP_NEGATIVEX = 0x800
+    CUBEMAP_POSITIVEY = 0x1000
+    CUBEMAP_NEGATIVEY = 0x2000
+    CUBEMAP_POSITIVEZ = 0x4000
+    CUBEMAP_NEGATIVEZ = 0x8000
+    VOLUME = 0x200000
+
+
+# Pixel Format
+class DDPF(IntFlag):
+    ALPHAPIXELS = 0x1
+    ALPHA = 0x2
+    FOURCC = 0x4
+    PALETTEINDEXED8 = 0x20
+    RGB = 0x40
+    LUMINANCE = 0x20000
+
+
+# dxgiformat.h
+class DXGI_FORMAT(IntEnum):
+    UNKNOWN = 0
+    R32G32B32A32_TYPELESS = 1
+    R32G32B32A32_FLOAT = 2
+    R32G32B32A32_UINT = 3
+    R32G32B32A32_SINT = 4
+    R32G32B32_TYPELESS = 5
+    R32G32B32_FLOAT = 6
+    R32G32B32_UINT = 7
+    R32G32B32_SINT = 8
+    R16G16B16A16_TYPELESS = 9
+    R16G16B16A16_FLOAT = 10
+    R16G16B16A16_UNORM = 11
+    R16G16B16A16_UINT = 12
+    R16G16B16A16_SNORM = 13
+    R16G16B16A16_SINT = 14
+    R32G32_TYPELESS = 15
+    R32G32_FLOAT = 16
+    R32G32_UINT = 17
+    R32G32_SINT = 18
+    R32G8X24_TYPELESS = 19
+    D32_FLOAT_S8X24_UINT = 20
+    R32_FLOAT_X8X24_TYPELESS = 21
+    X32_TYPELESS_G8X24_UINT = 22
+    R10G10B10A2_TYPELESS = 23
+    R10G10B10A2_UNORM = 24
+    R10G10B10A2_UINT = 25
+    R11G11B10_FLOAT = 26
+    R8G8B8A8_TYPELESS = 27
+    R8G8B8A8_UNORM = 28
+    R8G8B8A8_UNORM_SRGB = 29
+    R8G8B8A8_UINT = 30
+    R8G8B8A8_SNORM = 31
+    R8G8B8A8_SINT = 32
+    R16G16_TYPELESS = 33
+    R16G16_FLOAT = 34
+    R16G16_UNORM = 35
+    R16G16_UINT = 36
+    R16G16_SNORM = 37
+    R16G16_SINT = 38
+    R32_TYPELESS = 39
+    D32_FLOAT = 40
+    R32_FLOAT = 41
+    R32_UINT = 42
+    R32_SINT = 43
+    R24G8_TYPELESS = 44
+    D24_UNORM_S8_UINT = 45
+    R24_UNORM_X8_TYPELESS = 46
+    X24_TYPELESS_G8_UINT = 47
+    R8G8_TYPELESS = 48
+    R8G8_UNORM = 49
+    R8G8_UINT = 50
+    R8G8_SNORM = 51
+    R8G8_SINT = 52
+    R16_TYPELESS = 53
+    R16_FLOAT = 54
+    D16_UNORM = 55
+    R16_UNORM = 56
+    R16_UINT = 57
+    R16_SNORM = 58
+    R16_SINT = 59
+    R8_TYPELESS = 60
+    R8_UNORM = 61
+    R8_UINT = 62
+    R8_SNORM = 63
+    R8_SINT = 64
+    A8_UNORM = 65
+    R1_UNORM = 66
+    R9G9B9E5_SHAREDEXP = 67
+    R8G8_B8G8_UNORM = 68
+    G8R8_G8B8_UNORM = 69
+    BC1_TYPELESS = 70
+    BC1_UNORM = 71
+    BC1_UNORM_SRGB = 72
+    BC2_TYPELESS = 73
+    BC2_UNORM = 74
+    BC2_UNORM_SRGB = 75
+    BC3_TYPELESS = 76
+    BC3_UNORM = 77
+    BC3_UNORM_SRGB = 78
+    BC4_TYPELESS = 79
+    BC4_UNORM = 80
+    BC4_SNORM = 81
+    BC5_TYPELESS = 82
+    BC5_UNORM = 83
+    BC5_SNORM = 84
+    B5G6R5_UNORM = 85
+    B5G5R5A1_UNORM = 86
+    B8G8R8A8_UNORM = 87
+    B8G8R8X8_UNORM = 88
+    R10G10B10_XR_BIAS_A2_UNORM = 89
+    B8G8R8A8_TYPELESS = 90
+    B8G8R8A8_UNORM_SRGB = 91
+    B8G8R8X8_TYPELESS = 92
+    B8G8R8X8_UNORM_SRGB = 93
+    BC6H_TYPELESS = 94
+    BC6H_UF16 = 95
+    BC6H_SF16 = 96
+    BC7_TYPELESS = 97
+    BC7_UNORM = 98
+    BC7_UNORM_SRGB = 99
+    AYUV = 100
+    Y410 = 101
+    Y416 = 102
+    NV12 = 103
+    P010 = 104
+    P016 = 105
+    OPAQUE_420 = 106
+    YUY2 = 107
+    Y210 = 108
+    Y216 = 109
+    NV11 = 110
+    AI44 = 111
+    IA44 = 112
+    P8 = 113
+    A8P8 = 114
+    B4G4R4A4_UNORM = 115
+    P208 = 130
+    V208 = 131
+    V408 = 132
+    SAMPLER_FEEDBACK_MIN_MIP_OPAQUE = 189
+    SAMPLER_FEEDBACK_MIP_REGION_USED_OPAQUE = 190
+
+
+class D3DFMT(IntEnum):
+    UNKNOWN = 0
+    R8G8B8 = 20
+    A8R8G8B8 = 21
+    X8R8G8B8 = 22
+    R5G6B5 = 23
+    X1R5G5B5 = 24
+    A1R5G5B5 = 25
+    A4R4G4B4 = 26
+    R3G3B2 = 27
+    A8 = 28
+    A8R3G3B2 = 29
+    X4R4G4B4 = 30
+    A2B10G10R10 = 31
+    A8B8G8R8 = 32
+    X8B8G8R8 = 33
+    G16R16 = 34
+    A2R10G10B10 = 35
+    A16B16G16R16 = 36
+    A8P8 = 40
+    P8 = 41
+    L8 = 50
+    A8L8 = 51
+    A4L4 = 52
+    V8U8 = 60
+    L6V5U5 = 61
+    X8L8V8U8 = 62
+    Q8W8V8U8 = 63
+    V16U16 = 64
+    A2W10V10U10 = 67
+    D16_LOCKABLE = 70
+    D32 = 71
+    D15S1 = 73
+    D24S8 = 75
+    D24X8 = 77
+    D24X4S4 = 79
+    D16 = 80
+    D32F_LOCKABLE = 82
+    D24FS8 = 83
+    D32_LOCKABLE = 84
+    S8_LOCKABLE = 85
+    L16 = 81
+    VERTEXDATA = 100
+    INDEX16 = 101
+    INDEX32 = 102
+    Q16W16V16U16 = 110
+    R16F = 111
+    G16R16F = 112
+    A16B16G16R16F = 113
+    R32F = 114
+    G32R32F = 115
+    A32B32G32R32F = 116
+    CxV8U8 = 117
+    A1 = 118
+    A2B10G10R10_XR_BIAS = 119
+    BINARYBUFFER = 199
+
+    UYVY = i32(b"UYVY")
+    R8G8_B8G8 = i32(b"RGBG")
+    YUY2 = i32(b"YUY2")
+    G8R8_G8B8 = i32(b"GRGB")
+    DXT1 = i32(b"DXT1")
+    DXT2 = i32(b"DXT2")
+    DXT3 = i32(b"DXT3")
+    DXT4 = i32(b"DXT4")
+    DXT5 = i32(b"DXT5")
+    DX10 = i32(b"DX10")
+    BC4S = i32(b"BC4S")
+    BC4U = i32(b"BC4U")
+    BC5S = i32(b"BC5S")
+    BC5U = i32(b"BC5U")
+    ATI1 = i32(b"ATI1")
+    ATI2 = i32(b"ATI2")
+    MULTI2_ARGB8 = i32(b"MET1")
+
+
+# Backward compatibility layer
+module = sys.modules[__name__]
+for item in DDSD:
+    assert item.name is not None
+    setattr(module, f"DDSD_{item.name}", item.value)
+for item1 in DDSCAPS:
+    assert item1.name is not None
+    setattr(module, f"DDSCAPS_{item1.name}", item1.value)
+for item2 in DDSCAPS2:
+    assert item2.name is not None
+    setattr(module, f"DDSCAPS2_{item2.name}", item2.value)
+for item3 in DDPF:
+    assert item3.name is not None
+    setattr(module, f"DDPF_{item3.name}", item3.value)
+
+DDS_FOURCC = DDPF.FOURCC
+DDS_RGB = DDPF.RGB
+DDS_RGBA = DDPF.RGB | DDPF.ALPHAPIXELS
+DDS_LUMINANCE = DDPF.LUMINANCE
+DDS_LUMINANCEA = DDPF.LUMINANCE | DDPF.ALPHAPIXELS
+DDS_ALPHA = DDPF.ALPHA
+DDS_PAL8 = DDPF.PALETTEINDEXED8
+
+DDS_HEADER_FLAGS_TEXTURE = DDSD.CAPS | DDSD.HEIGHT | DDSD.WIDTH | DDSD.PIXELFORMAT
+DDS_HEADER_FLAGS_MIPMAP = DDSD.MIPMAPCOUNT
+DDS_HEADER_FLAGS_VOLUME = DDSD.DEPTH
+DDS_HEADER_FLAGS_PITCH = DDSD.PITCH
+DDS_HEADER_FLAGS_LINEARSIZE = DDSD.LINEARSIZE
+
+DDS_HEIGHT = DDSD.HEIGHT
+DDS_WIDTH = DDSD.WIDTH
+
+DDS_SURFACE_FLAGS_TEXTURE = DDSCAPS.TEXTURE
+DDS_SURFACE_FLAGS_MIPMAP = DDSCAPS.COMPLEX | DDSCAPS.MIPMAP
+DDS_SURFACE_FLAGS_CUBEMAP = DDSCAPS.COMPLEX
+
+DDS_CUBEMAP_POSITIVEX = DDSCAPS2.CUBEMAP | DDSCAPS2.CUBEMAP_POSITIVEX
+DDS_CUBEMAP_NEGATIVEX = DDSCAPS2.CUBEMAP | DDSCAPS2.CUBEMAP_NEGATIVEX
+DDS_CUBEMAP_POSITIVEY = DDSCAPS2.CUBEMAP | DDSCAPS2.CUBEMAP_POSITIVEY
+DDS_CUBEMAP_NEGATIVEY = DDSCAPS2.CUBEMAP | DDSCAPS2.CUBEMAP_NEGATIVEY
+DDS_CUBEMAP_POSITIVEZ = DDSCAPS2.CUBEMAP | DDSCAPS2.CUBEMAP_POSITIVEZ
+DDS_CUBEMAP_NEGATIVEZ = DDSCAPS2.CUBEMAP | DDSCAPS2.CUBEMAP_NEGATIVEZ
+
+DXT1_FOURCC = D3DFMT.DXT1
+DXT3_FOURCC = D3DFMT.DXT3
+DXT5_FOURCC = D3DFMT.DXT5
+
+DXGI_FORMAT_R8G8B8A8_TYPELESS = DXGI_FORMAT.R8G8B8A8_TYPELESS
+DXGI_FORMAT_R8G8B8A8_UNORM = DXGI_FORMAT.R8G8B8A8_UNORM
+DXGI_FORMAT_R8G8B8A8_UNORM_SRGB = DXGI_FORMAT.R8G8B8A8_UNORM_SRGB
+DXGI_FORMAT_BC5_TYPELESS = DXGI_FORMAT.BC5_TYPELESS
+DXGI_FORMAT_BC5_UNORM = DXGI_FORMAT.BC5_UNORM
+DXGI_FORMAT_BC5_SNORM = DXGI_FORMAT.BC5_SNORM
+DXGI_FORMAT_BC6H_UF16 = DXGI_FORMAT.BC6H_UF16
+DXGI_FORMAT_BC6H_SF16 = DXGI_FORMAT.BC6H_SF16
+DXGI_FORMAT_BC7_TYPELESS = DXGI_FORMAT.BC7_TYPELESS
+DXGI_FORMAT_BC7_UNORM = DXGI_FORMAT.BC7_UNORM
+DXGI_FORMAT_BC7_UNORM_SRGB = DXGI_FORMAT.BC7_UNORM_SRGB
+
+
+class DdsImageFile(ImageFile.ImageFile):
+    format = "DDS"
+    format_description = "DirectDraw Surface"
+
+    def _open(self) -> None:
+        if not _accept(self.fp.read(4)):
+            msg = "not a DDS file"
+            raise SyntaxError(msg)
+        (header_size,) = struct.unpack("<I", self.fp.read(4))
+        if header_size != 124:
+            msg = f"Unsupported header size {repr(header_size)}"
+            raise OSError(msg)
+        header_bytes = self.fp.read(header_size - 4)
+        if len(header_bytes) != 120:
+            msg = f"Incomplete header: {len(header_bytes)} bytes"
+            raise OSError(msg)
+        header = io.BytesIO(header_bytes)
+
+        flags, height, width = struct.unpack("<3I", header.read(12))
+        self._size = (width, height)
+        extents = (0, 0) + self.size
+
+        pitch, depth, mipmaps = struct.unpack("<3I", header.read(12))
+        struct.unpack("<11I", header.read(44))  # reserved
+
+        # pixel format
+        pfsize, pfflags, fourcc, bitcount = struct.unpack("<4I", header.read(16))
+        n = 0
+        rawmode = None
+        if pfflags & DDPF.RGB:
+            # Texture contains uncompressed RGB data
+            if pfflags & DDPF.ALPHAPIXELS:
+                self._mode = "RGBA"
+                mask_count = 4
+            else:
+                self._mode = "RGB"
+                mask_count = 3
+
+            masks = struct.unpack(f"<{mask_count}I", header.read(mask_count * 4))
+            self.tile = [ImageFile._Tile("dds_rgb", extents, 0, (bitcount, masks))]
+            return
+        elif pfflags & DDPF.LUMINANCE:
+            if bitcount == 8:
+                self._mode = "L"
+            elif bitcount == 16 and pfflags & DDPF.ALPHAPIXELS:
+                self._mode = "LA"
+            else:
+                msg = f"Unsupported bitcount {bitcount} for {pfflags}"
+                raise OSError(msg)
+        elif pfflags & DDPF.PALETTEINDEXED8:
+            self._mode = "P"
+            self.palette = ImagePalette.raw("RGBA", self.fp.read(1024))
+            self.palette.mode = "RGBA"
+        elif pfflags & DDPF.FOURCC:
+            offset = header_size + 4
+            if fourcc == D3DFMT.DXT1:
+                self._mode = "RGBA"
+                self.pixel_format = "DXT1"
+                n = 1
+            elif fourcc == D3DFMT.DXT3:
+                self._mode = "RGBA"
+                self.pixel_format = "DXT3"
+                n = 2
+            elif fourcc == D3DFMT.DXT5:
+                self._mode = "RGBA"
+                self.pixel_format = "DXT5"
+                n = 3
+            elif fourcc in (D3DFMT.BC4U, D3DFMT.ATI1):
+                self._mode = "L"
+                self.pixel_format = "BC4"
+                n = 4
+            elif fourcc == D3DFMT.BC5S:
+                self._mode = "RGB"
+                self.pixel_format = "BC5S"
+                n = 5
+            elif fourcc in (D3DFMT.BC5U, D3DFMT.ATI2):
+                self._mode = "RGB"
+                self.pixel_format = "BC5"
+                n = 5
+            elif fourcc == D3DFMT.DX10:
+                offset += 20
+                # ignoring flags which pertain to volume textures and cubemaps
+                (dxgi_format,) = struct.unpack("<I", self.fp.read(4))
+                self.fp.read(16)
+                if dxgi_format in (
+                    DXGI_FORMAT.BC1_UNORM,
+                    DXGI_FORMAT.BC1_TYPELESS,
+                ):
+                    self._mode = "RGBA"
+                    self.pixel_format = "BC1"
+                    n = 1
+                elif dxgi_format in (DXGI_FORMAT.BC4_TYPELESS, DXGI_FORMAT.BC4_UNORM):
+                    self._mode = "L"
+                    self.pixel_format = "BC4"
+                    n = 4
+                elif dxgi_format in (DXGI_FORMAT.BC5_TYPELESS, DXGI_FORMAT.BC5_UNORM):
+                    self._mode = "RGB"
+                    self.pixel_format = "BC5"
+                    n = 5
+                elif dxgi_format == DXGI_FORMAT.BC5_SNORM:
+                    self._mode = "RGB"
+                    self.pixel_format = "BC5S"
+                    n = 5
+                elif dxgi_format == DXGI_FORMAT.BC6H_UF16:
+                    self._mode = "RGB"
+                    self.pixel_format = "BC6H"
+                    n = 6
+                elif dxgi_format == DXGI_FORMAT.BC6H_SF16:
+                    self._mode = "RGB"
+                    self.pixel_format = "BC6HS"
+                    n = 6
+                elif dxgi_format in (
+                    DXGI_FORMAT.BC7_TYPELESS,
+                    DXGI_FORMAT.BC7_UNORM,
+                    DXGI_FORMAT.BC7_UNORM_SRGB,
+                ):
+                    self._mode = "RGBA"
+                    self.pixel_format = "BC7"
+                    n = 7
+                    if dxgi_format == DXGI_FORMAT.BC7_UNORM_SRGB:
+                        self.info["gamma"] = 1 / 2.2
+                elif dxgi_format in (
+                    DXGI_FORMAT.R8G8B8A8_TYPELESS,
+                    DXGI_FORMAT.R8G8B8A8_UNORM,
+                    DXGI_FORMAT.R8G8B8A8_UNORM_SRGB,
+                ):
+                    self._mode = "RGBA"
+                    if dxgi_format == DXGI_FORMAT.R8G8B8A8_UNORM_SRGB:
+                        self.info["gamma"] = 1 / 2.2
+                else:
+                    msg = f"Unimplemented DXGI format {dxgi_format}"
+                    raise NotImplementedError(msg)
+            else:
+                msg = f"Unimplemented pixel format {repr(fourcc)}"
+                raise NotImplementedError(msg)
+        else:
+            msg = f"Unknown pixel format flags {pfflags}"
+            raise NotImplementedError(msg)
+
+        if n:
+            self.tile = [
+                ImageFile._Tile("bcn", extents, offset, (n, self.pixel_format))
+            ]
+        else:
+            self.tile = [ImageFile._Tile("raw", extents, 0, rawmode or self.mode)]
+
+    def load_seek(self, pos: int) -> None:
+        pass
+
+
+class DdsRgbDecoder(ImageFile.PyDecoder):
+    _pulls_fd = True
+
+    def decode(self, buffer: bytes | Image.SupportsArrayInterface) -> tuple[int, int]:
+        assert self.fd is not None
+        bitcount, masks = self.args
+
+        # Some masks will be padded with zeros, e.g. R 0b11 G 0b1100
+        # Calculate how many zeros each mask is padded with
+        mask_offsets = []
+        # And the maximum value of each channel without the padding
+        mask_totals = []
+        for mask in masks:
+            offset = 0
+            if mask != 0:
+                while mask >> (offset + 1) << (offset + 1) == mask:
+                    offset += 1
+            mask_offsets.append(offset)
+            mask_totals.append(mask >> offset)
+
+        data = bytearray()
+        bytecount = bitcount // 8
+        dest_length = self.state.xsize * self.state.ysize * len(masks)
+        while len(data) < dest_length:
+            value = int.from_bytes(self.fd.read(bytecount), "little")
+            for i, mask in enumerate(masks):
+                masked_value = value & mask
+                # Remove the zero padding, and scale it to 8 bits
+                data += o8(
+                    int(((masked_value >> mask_offsets[i]) / mask_totals[i]) * 255)
+                )
+        self.set_as_raw(data)
+        return -1, 0
+
+
+def _save(im: Image.Image, fp: IO[bytes], filename: str | bytes) -> None:
+    if im.mode not in ("RGB", "RGBA", "L", "LA"):
+        msg = f"cannot write mode {im.mode} as DDS"
+        raise OSError(msg)
+
+    alpha = im.mode[-1] == "A"
+    if im.mode[0] == "L":
+        pixel_flags = DDPF.LUMINANCE
+        rawmode = im.mode
+        if alpha:
+            rgba_mask = [0x000000FF, 0x000000FF, 0x000000FF]
+        else:
+            rgba_mask = [0xFF000000, 0xFF000000, 0xFF000000]
+    else:
+        pixel_flags = DDPF.RGB
+        rawmode = im.mode[::-1]
+        rgba_mask = [0x00FF0000, 0x0000FF00, 0x000000FF]
+
+        if alpha:
+            r, g, b, a = im.split()
+            im = Image.merge("RGBA", (a, r, g, b))
+    if alpha:
+        pixel_flags |= DDPF.ALPHAPIXELS
+    rgba_mask.append(0xFF000000 if alpha else 0)
+
+    flags = DDSD.CAPS | DDSD.HEIGHT | DDSD.WIDTH | DDSD.PITCH | DDSD.PIXELFORMAT
+    bitcount = len(im.getbands()) * 8
+    pitch = (im.width * bitcount + 7) // 8
+
+    fp.write(
+        o32(DDS_MAGIC)
+        + struct.pack(
+            "<7I",
+            124,  # header size
+            flags,  # flags
+            im.height,
+            im.width,
+            pitch,
+            0,  # depth
+            0,  # mipmaps
+        )
+        + struct.pack("11I", *((0,) * 11))  # reserved
+        # pfsize, pfflags, fourcc, bitcount
+        + struct.pack("<4I", 32, pixel_flags, 0, bitcount)
+        + struct.pack("<4I", *rgba_mask)  # dwRGBABitMask
+        + struct.pack("<5I", DDSCAPS.TEXTURE, 0, 0, 0, 0)
+    )
+    ImageFile._save(im, fp, [ImageFile._Tile("raw", (0, 0) + im.size, 0, rawmode)])
+
+
+def _accept(prefix: bytes) -> bool:
+    return prefix[:4] == b"DDS "
+
+
+Image.register_open(DdsImageFile.format, DdsImageFile, _accept)
+Image.register_decoder("dds_rgb", DdsRgbDecoder)
+Image.register_save(DdsImageFile.format, _save)
+Image.register_extension(DdsImageFile.format, ".dds")

venv/Lib/site-packages/PIL/EpsImagePlugin.py

@@ -0,0 +1,474 @@
+#
+# The Python Imaging Library.
+# $Id$
+#
+# EPS file handling
+#
+# History:
+# 1995-09-01 fl   Created (0.1)
+# 1996-05-18 fl   Don't choke on "atend" fields, Ghostscript interface (0.2)
+# 1996-08-22 fl   Don't choke on floating point BoundingBox values
+# 1996-08-23 fl   Handle files from Macintosh (0.3)
+# 2001-02-17 fl   Use 're' instead of 'regex' (Python 2.1) (0.4)
+# 2003-09-07 fl   Check gs.close status (from Federico Di Gregorio) (0.5)
+# 2014-05-07 e    Handling of EPS with binary preview and fixed resolution
+#                 resizing
+#
+# Copyright (c) 1997-2003 by Secret Labs AB.
+# Copyright (c) 1995-2003 by Fredrik Lundh
+#
+# See the README file for information on usage and redistribution.
+#
+from __future__ import annotations
+
+import io
+import os
+import re
+import subprocess
+import sys
+import tempfile
+from typing import IO
+
+from . import Image, ImageFile
+from ._binary import i32le as i32
+
+# --------------------------------------------------------------------
+
+
+split = re.compile(r"^%%([^:]*):[ \t]*(.*)[ \t]*$")
+field = re.compile(r"^%[%!\w]([^:]*)[ \t]*$")
+
+gs_binary: str | bool | None = None
+gs_windows_binary = None
+
+
+def has_ghostscript() -> bool:
+    global gs_binary, gs_windows_binary
+    if gs_binary is None:
+        if sys.platform.startswith("win"):
+            if gs_windows_binary is None:
+                import shutil
+
+                for binary in ("gswin32c", "gswin64c", "gs"):
+                    if shutil.which(binary) is not None:
+                        gs_windows_binary = binary
+                        break
+                else:
+                    gs_windows_binary = False
+            gs_binary = gs_windows_binary
+        else:
+            try:
+                subprocess.check_call(["gs", "--version"], stdout=subprocess.DEVNULL)
+                gs_binary = "gs"
+            except OSError:
+                gs_binary = False
+    return gs_binary is not False
+
+
+def Ghostscript(
+    tile: list[ImageFile._Tile],
+    size: tuple[int, int],
+    fp: IO[bytes],
+    scale: int = 1,
+    transparency: bool = False,
+) -> Image.core.ImagingCore:
+    """Render an image using Ghostscript"""
+    global gs_binary
+    if not has_ghostscript():
+        msg = "Unable to locate Ghostscript on paths"
+        raise OSError(msg)
+    assert isinstance(gs_binary, str)
+
+    # Unpack decoder tile
+    args = tile[0].args
+    assert isinstance(args, tuple)
+    length, bbox = args
+
+    # Hack to support hi-res rendering
+    scale = int(scale) or 1
+    width = size[0] * scale
+    height = size[1] * scale
+    # resolution is dependent on bbox and size
+    res_x = 72.0 * width / (bbox[2] - bbox[0])
+    res_y = 72.0 * height / (bbox[3] - bbox[1])
+
+    out_fd, outfile = tempfile.mkstemp()
+    os.close(out_fd)
+
+    infile_temp = None
+    if hasattr(fp, "name") and os.path.exists(fp.name):
+        infile = fp.name
+    else:
+        in_fd, infile_temp = tempfile.mkstemp()
+        os.close(in_fd)
+        infile = infile_temp
+
+        # Ignore length and offset!
+        # Ghostscript can read it
+        # Copy whole file to read in Ghostscript
+        with open(infile_temp, "wb") as f:
+            # fetch length of fp
+            fp.seek(0, io.SEEK_END)
+            fsize = fp.tell()
+            # ensure start position
+            # go back
+            fp.seek(0)
+            lengthfile = fsize
+            while lengthfile > 0:
+                s = fp.read(min(lengthfile, 100 * 1024))
+                if not s:
+                    break
+                lengthfile -= len(s)
+                f.write(s)
+
+    if transparency:
+        # "RGBA"
+        device = "pngalpha"
+    else:
+        # "pnmraw" automatically chooses between
+        # PBM ("1"), PGM ("L"), and PPM ("RGB").
+        device = "pnmraw"
+
+    # Build Ghostscript command
+    command = [
+        gs_binary,
+        "-q",  # quiet mode
+        f"-g{width:d}x{height:d}",  # set output geometry (pixels)
+        f"-r{res_x:f}x{res_y:f}",  # set input DPI (dots per inch)
+        "-dBATCH",  # exit after processing
+        "-dNOPAUSE",  # don't pause between pages
+        "-dSAFER",  # safe mode
+        f"-sDEVICE={device}",
+        f"-sOutputFile={outfile}",  # output file
+        # adjust for image origin
+        "-c",
+        f"{-bbox[0]} {-bbox[1]} translate",
+        "-f",
+        infile,  # input file
+        # showpage (see https://bugs.ghostscript.com/show_bug.cgi?id=698272)
+        "-c",
+        "showpage",
+    ]
+
+    # push data through Ghostscript
+    try:
+        startupinfo = None
+        if sys.platform.startswith("win"):
+            startupinfo = subprocess.STARTUPINFO()
+            startupinfo.dwFlags |= subprocess.STARTF_USESHOWWINDOW
+        subprocess.check_call(command, startupinfo=startupinfo)
+        with Image.open(outfile) as out_im:
+            out_im.load()
+            return out_im.im.copy()
+    finally:
+        try:
+            os.unlink(outfile)
+            if infile_temp:
+                os.unlink(infile_temp)
+        except OSError:
+            pass
+
+
+def _accept(prefix: bytes) -> bool:
+    return prefix[:4] == b"%!PS" or (len(prefix) >= 4 and i32(prefix) == 0xC6D3D0C5)
+
+
+##
+# Image plugin for Encapsulated PostScript. This plugin supports only
+# a few variants of this format.
+
+
+class EpsImageFile(ImageFile.ImageFile):
+    """EPS File Parser for the Python Imaging Library"""
+
+    format = "EPS"
+    format_description = "Encapsulated Postscript"
+
+    mode_map = {1: "L", 2: "LAB", 3: "RGB", 4: "CMYK"}
+
+    def _open(self) -> None:
+        (length, offset) = self._find_offset(self.fp)
+
+        # go to offset - start of "%!PS"
+        self.fp.seek(offset)
+
+        self._mode = "RGB"
+
+        # When reading header comments, the first comment is used.
+        # When reading trailer comments, the last comment is used.
+        bounding_box: list[int] | None = None
+        imagedata_size: tuple[int, int] | None = None
+
+        byte_arr = bytearray(255)
+        bytes_mv = memoryview(byte_arr)
+        bytes_read = 0
+        reading_header_comments = True
+        reading_trailer_comments = False
+        trailer_reached = False
+
+        def check_required_header_comments() -> None:
+            """
+            The EPS specification requires that some headers exist.
+            This should be checked when the header comments formally end,
+            when image data starts, or when the file ends, whichever comes first.
+            """
+            if "PS-Adobe" not in self.info:
+                msg = 'EPS header missing "%!PS-Adobe" comment'
+                raise SyntaxError(msg)
+            if "BoundingBox" not in self.info:
+                msg = 'EPS header missing "%%BoundingBox" comment'
+                raise SyntaxError(msg)
+
+        def read_comment(s: str) -> bool:
+            nonlocal bounding_box, reading_trailer_comments
+            try:
+                m = split.match(s)
+            except re.error as e:
+                msg = "not an EPS file"
+                raise SyntaxError(msg) from e
+
+            if not m:
+                return False
+
+            k, v = m.group(1, 2)
+            self.info[k] = v
+            if k == "BoundingBox":
+                if v == "(atend)":
+                    reading_trailer_comments = True
+                elif not bounding_box or (trailer_reached and reading_trailer_comments):
+                    try:
+                        # Note: The DSC spec says that BoundingBox
+                        # fields should be integers, but some drivers
+                        # put floating point values there anyway.
+                        bounding_box = [int(float(i)) for i in v.split()]
+                    except Exception:
+                        pass
+            return True
+
+        while True:
+            byte = self.fp.read(1)
+            if byte == b"":
+                # if we didn't read a byte we must be at the end of the file
+                if bytes_read == 0:
+                    if reading_header_comments:
+                        check_required_header_comments()
+                    break
+            elif byte in b"\r\n":
+                # if we read a line ending character, ignore it and parse what
+                # we have already read. if we haven't read any other characters,
+                # continue reading
+                if bytes_read == 0:
+                    continue
+            else:
+                # ASCII/hexadecimal lines in an EPS file must not exceed
+                # 255 characters, not including line ending characters
+                if bytes_read >= 255:
+                    # only enforce this for lines starting with a "%",
+                    # otherwise assume it's binary data
+                    if byte_arr[0] == ord("%"):
+                        msg = "not an EPS file"
+                        raise SyntaxError(msg)
+                    else:
+                        if reading_header_comments:
+                            check_required_header_comments()
+                            reading_header_comments = False
+                        # reset bytes_read so we can keep reading
+                        # data until the end of the line
+                        bytes_read = 0
+                byte_arr[bytes_read] = byte[0]
+                bytes_read += 1
+                continue
+
+            if reading_header_comments:
+                # Load EPS header
+
+                # if this line doesn't start with a "%",
+                # or does start with "%%EndComments",
+                # then we've reached the end of the header/comments
+                if byte_arr[0] != ord("%") or bytes_mv[:13] == b"%%EndComments":
+                    check_required_header_comments()
+                    reading_header_comments = False
+                    continue
+
+                s = str(bytes_mv[:bytes_read], "latin-1")
+                if not read_comment(s):
+                    m = field.match(s)
+                    if m:
+                        k = m.group(1)
+                        if k[:8] == "PS-Adobe":
+                            self.info["PS-Adobe"] = k[9:]
+                        else:
+                            self.info[k] = ""
+                    elif s[0] == "%":
+                        # handle non-DSC PostScript comments that some
+                        # tools mistakenly put in the Comments section
+                        pass
+                    else:
+                        msg = "bad EPS header"
+                        raise OSError(msg)
+            elif bytes_mv[:11] == b"%ImageData:":
+                # Check for an "ImageData" descriptor
+                # https://www.adobe.com/devnet-apps/photoshop/fileformatashtml/#50577413_pgfId-1035096
+
+                # If we've already read an "ImageData" descriptor,
+                # don't read another one.
+                if imagedata_size:
+                    bytes_read = 0
+                    continue
+
+                # Values:
+                # columns
+                # rows
+                # bit depth (1 or 8)
+                # mode (1: L, 2: LAB, 3: RGB, 4: CMYK)
+                # number of padding channels
+                # block size (number of bytes per row per channel)
+                # binary/ascii (1: binary, 2: ascii)
+                # data start identifier (the image data follows after a single line
+                #   consisting only of this quoted value)
+                image_data_values = byte_arr[11:bytes_read].split(None, 7)
+                columns, rows, bit_depth, mode_id = (
+                    int(value) for value in image_data_values[:4]
+                )
+
+                if bit_depth == 1:
+                    self._mode = "1"
+                elif bit_depth == 8:
+                    try:
+                        self._mode = self.mode_map[mode_id]
+                    except ValueError:
+                        break
+                else:
+                    break
+
+                # Parse the columns and rows after checking the bit depth and mode
+                # in case the bit depth and/or mode are invalid.
+                imagedata_size = columns, rows
+            elif bytes_mv[:5] == b"%%EOF":
+                break
+            elif trailer_reached and reading_trailer_comments:
+                # Load EPS trailer
+                s = str(bytes_mv[:bytes_read], "latin-1")
+                read_comment(s)
+            elif bytes_mv[:9] == b"%%Trailer":
+                trailer_reached = True
+            bytes_read = 0
+
+        # A "BoundingBox" is always required,
+        # even if an "ImageData" descriptor size exists.
+        if not bounding_box:
+            msg = "cannot determine EPS bounding box"
+            raise OSError(msg)
+
+        # An "ImageData" size takes precedence over the "BoundingBox".
+        self._size = imagedata_size or (
+            bounding_box[2] - bounding_box[0],
+            bounding_box[3] - bounding_box[1],
+        )
+
+        self.tile = [
+            ImageFile._Tile("eps", (0, 0) + self.size, offset, (length, bounding_box))
+        ]
+
+    def _find_offset(self, fp: IO[bytes]) -> tuple[int, int]:
+        s = fp.read(4)
+
+        if s == b"%!PS":
+            # for HEAD without binary preview
+            fp.seek(0, io.SEEK_END)
+            length = fp.tell()
+            offset = 0
+        elif i32(s) == 0xC6D3D0C5:
+            # FIX for: Some EPS file not handled correctly / issue #302
+            # EPS can contain binary data
+            # or start directly with latin coding
+            # more info see:
+            # https://web.archive.org/web/20160528181353/http://partners.adobe.com/public/developer/en/ps/5002.EPSF_Spec.pdf
+            s = fp.read(8)
+            offset = i32(s)
+            length = i32(s, 4)
+        else:
+            msg = "not an EPS file"
+            raise SyntaxError(msg)
+
+        return length, offset
+
+    def load(
+        self, scale: int = 1, transparency: bool = False
+    ) -> Image.core.PixelAccess | None:
+        # Load EPS via Ghostscript
+        if self.tile:
+            self.im = Ghostscript(self.tile, self.size, self.fp, scale, transparency)
+            self._mode = self.im.mode
+            self._size = self.im.size
+            self.tile = []
+        return Image.Image.load(self)
+
+    def load_seek(self, pos: int) -> None:
+        # we can't incrementally load, so force ImageFile.parser to
+        # use our custom load method by defining this method.
+        pass
+
+
+# --------------------------------------------------------------------
+
+
+def _save(im: Image.Image, fp: IO[bytes], filename: str | bytes, eps: int = 1) -> None:
+    """EPS Writer for the Python Imaging Library."""
+
+    # make sure image data is available
+    im.load()
+
+    # determine PostScript image mode
+    if im.mode == "L":
+        operator = (8, 1, b"image")
+    elif im.mode == "RGB":
+        operator = (8, 3, b"false 3 colorimage")
+    elif im.mode == "CMYK":
+        operator = (8, 4, b"false 4 colorimage")
+    else:
+        msg = "image mode is not supported"
+        raise ValueError(msg)
+
+    if eps:
+        # write EPS header
+        fp.write(b"%!PS-Adobe-3.0 EPSF-3.0\n")
+        fp.write(b"%%Creator: PIL 0.1 EpsEncode\n")
+        # fp.write("%%CreationDate: %s"...)
+        fp.write(b"%%%%BoundingBox: 0 0 %d %d\n" % im.size)
+        fp.write(b"%%Pages: 1\n")
+        fp.write(b"%%EndComments\n")
+        fp.write(b"%%Page: 1 1\n")
+        fp.write(b"%%ImageData: %d %d " % im.size)
+        fp.write(b'%d %d 0 1 1 "%s"\n' % operator)
+
+    # image header
+    fp.write(b"gsave\n")
+    fp.write(b"10 dict begin\n")
+    fp.write(b"/buf %d string def\n" % (im.size[0] * operator[1]))
+    fp.write(b"%d %d scale\n" % im.size)
+    fp.write(b"%d %d 8\n" % im.size)  # <= bits
+    fp.write(b"[%d 0 0 -%d 0 %d]\n" % (im.size[0], im.size[1], im.size[1]))
+    fp.write(b"{ currentfile buf readhexstring pop } bind\n")
+    fp.write(operator[2] + b"\n")
+    if hasattr(fp, "flush"):
+        fp.flush()
+
+    ImageFile._save(im, fp, [ImageFile._Tile("eps", (0, 0) + im.size)])
+
+    fp.write(b"\n%%%%EndBinary\n")
+    fp.write(b"grestore end\n")
+    if hasattr(fp, "flush"):
+        fp.flush()
+
+
+# --------------------------------------------------------------------
+
+
+Image.register_open(EpsImageFile.format, EpsImageFile, _accept)
+
+Image.register_save(EpsImageFile.format, _save)
+
+Image.register_extensions(EpsImageFile.format, [".ps", ".eps"])
+
+Image.register_mime(EpsImageFile.format, "application/postscript")

venv/Lib/site-packages/PIL/ExifTags.py

@@ -0,0 +1,382 @@
+#
+# The Python Imaging Library.
+# $Id$
+#
+# EXIF tags
+#
+# Copyright (c) 2003 by Secret Labs AB
+#
+# See the README file for information on usage and redistribution.
+#
+
+"""
+This module provides constants and clear-text names for various
+well-known EXIF tags.
+"""
+from __future__ import annotations
+
+from enum import IntEnum
+
+
+class Base(IntEnum):
+    # possibly incomplete
+    InteropIndex = 0x0001
+    ProcessingSoftware = 0x000B
+    NewSubfileType = 0x00FE
+    SubfileType = 0x00FF
+    ImageWidth = 0x0100
+    ImageLength = 0x0101
+    BitsPerSample = 0x0102
+    Compression = 0x0103
+    PhotometricInterpretation = 0x0106
+    Thresholding = 0x0107
+    CellWidth = 0x0108
+    CellLength = 0x0109
+    FillOrder = 0x010A
+    DocumentName = 0x010D
+    ImageDescription = 0x010E
+    Make = 0x010F
+    Model = 0x0110
+    StripOffsets = 0x0111
+    Orientation = 0x0112
+    SamplesPerPixel = 0x0115
+    RowsPerStrip = 0x0116
+    StripByteCounts = 0x0117
+    MinSampleValue = 0x0118
+    MaxSampleValue = 0x0119
+    XResolution = 0x011A
+    YResolution = 0x011B
+    PlanarConfiguration = 0x011C
+    PageName = 0x011D
+    FreeOffsets = 0x0120
+    FreeByteCounts = 0x0121
+    GrayResponseUnit = 0x0122
+    GrayResponseCurve = 0x0123
+    T4Options = 0x0124
+    T6Options = 0x0125
+    ResolutionUnit = 0x0128
+    PageNumber = 0x0129
+    TransferFunction = 0x012D
+    Software = 0x0131
+    DateTime = 0x0132
+    Artist = 0x013B
+    HostComputer = 0x013C
+    Predictor = 0x013D
+    WhitePoint = 0x013E
+    PrimaryChromaticities = 0x013F
+    ColorMap = 0x0140
+    HalftoneHints = 0x0141
+    TileWidth = 0x0142
+    TileLength = 0x0143
+    TileOffsets = 0x0144
+    TileByteCounts = 0x0145
+    SubIFDs = 0x014A
+    InkSet = 0x014C
+    InkNames = 0x014D
+    NumberOfInks = 0x014E
+    DotRange = 0x0150
+    TargetPrinter = 0x0151
+    ExtraSamples = 0x0152
+    SampleFormat = 0x0153
+    SMinSampleValue = 0x0154
+    SMaxSampleValue = 0x0155
+    TransferRange = 0x0156
+    ClipPath = 0x0157
+    XClipPathUnits = 0x0158
+    YClipPathUnits = 0x0159
+    Indexed = 0x015A
+    JPEGTables = 0x015B
+    OPIProxy = 0x015F
+    JPEGProc = 0x0200
+    JpegIFOffset = 0x0201
+    JpegIFByteCount = 0x0202
+    JpegRestartInterval = 0x0203
+    JpegLosslessPredictors = 0x0205
+    JpegPointTransforms = 0x0206
+    JpegQTables = 0x0207
+    JpegDCTables = 0x0208
+    JpegACTables = 0x0209
+    YCbCrCoefficients = 0x0211
+    YCbCrSubSampling = 0x0212
+    YCbCrPositioning = 0x0213
+    ReferenceBlackWhite = 0x0214
+    XMLPacket = 0x02BC
+    RelatedImageFileFormat = 0x1000
+    RelatedImageWidth = 0x1001
+    RelatedImageLength = 0x1002
+    Rating = 0x4746
+    RatingPercent = 0x4749
+    ImageID = 0x800D
+    CFARepeatPatternDim = 0x828D
+    BatteryLevel = 0x828F
+    Copyright = 0x8298
+    ExposureTime = 0x829A
+    FNumber = 0x829D
+    IPTCNAA = 0x83BB
+    ImageResources = 0x8649
+    ExifOffset = 0x8769
+    InterColorProfile = 0x8773
+    ExposureProgram = 0x8822
+    SpectralSensitivity = 0x8824
+    GPSInfo = 0x8825
+    ISOSpeedRatings = 0x8827
+    OECF = 0x8828
+    Interlace = 0x8829
+    TimeZoneOffset = 0x882A
+    SelfTimerMode = 0x882B
+    SensitivityType = 0x8830
+    StandardOutputSensitivity = 0x8831
+    RecommendedExposureIndex = 0x8832
+    ISOSpeed = 0x8833
+    ISOSpeedLatitudeyyy = 0x8834
+    ISOSpeedLatitudezzz = 0x8835
+    ExifVersion = 0x9000
+    DateTimeOriginal = 0x9003
+    DateTimeDigitized = 0x9004
+    OffsetTime = 0x9010
+    OffsetTimeOriginal = 0x9011
+    OffsetTimeDigitized = 0x9012
+    ComponentsConfiguration = 0x9101
+    CompressedBitsPerPixel = 0x9102
+    ShutterSpeedValue = 0x9201
+    ApertureValue = 0x9202
+    BrightnessValue = 0x9203
+    ExposureBiasValue = 0x9204
+    MaxApertureValue = 0x9205
+    SubjectDistance = 0x9206
+    MeteringMode = 0x9207
+    LightSource = 0x9208
+    Flash = 0x9209
+    FocalLength = 0x920A
+    Noise = 0x920D
+    ImageNumber = 0x9211
+    SecurityClassification = 0x9212
+    ImageHistory = 0x9213
+    TIFFEPStandardID = 0x9216
+    MakerNote = 0x927C
+    UserComment = 0x9286
+    SubsecTime = 0x9290
+    SubsecTimeOriginal = 0x9291
+    SubsecTimeDigitized = 0x9292
+    AmbientTemperature = 0x9400
+    Humidity = 0x9401
+    Pressure = 0x9402
+    WaterDepth = 0x9403
+    Acceleration = 0x9404
+    CameraElevationAngle = 0x9405
+    XPTitle = 0x9C9B
+    XPComment = 0x9C9C
+    XPAuthor = 0x9C9D
+    XPKeywords = 0x9C9E
+    XPSubject = 0x9C9F
+    FlashPixVersion = 0xA000
+    ColorSpace = 0xA001
+    ExifImageWidth = 0xA002
+    ExifImageHeight = 0xA003
+    RelatedSoundFile = 0xA004
+    ExifInteroperabilityOffset = 0xA005
+    FlashEnergy = 0xA20B
+    SpatialFrequencyResponse = 0xA20C
+    FocalPlaneXResolution = 0xA20E
+    FocalPlaneYResolution = 0xA20F
+    FocalPlaneResolutionUnit = 0xA210
+    SubjectLocation = 0xA214
+    ExposureIndex = 0xA215
+    SensingMethod = 0xA217
+    FileSource = 0xA300
+    SceneType = 0xA301
+    CFAPattern = 0xA302
+    CustomRendered = 0xA401
+    ExposureMode = 0xA402
+    WhiteBalance = 0xA403
+    DigitalZoomRatio = 0xA404
+    FocalLengthIn35mmFilm = 0xA405
+    SceneCaptureType = 0xA406
+    GainControl = 0xA407
+    Contrast = 0xA408
+    Saturation = 0xA409
+    Sharpness = 0xA40A
+    DeviceSettingDescription = 0xA40B
+    SubjectDistanceRange = 0xA40C
+    ImageUniqueID = 0xA420
+    CameraOwnerName = 0xA430
+    BodySerialNumber = 0xA431
+    LensSpecification = 0xA432
+    LensMake = 0xA433
+    LensModel = 0xA434
+    LensSerialNumber = 0xA435
+    CompositeImage = 0xA460
+    CompositeImageCount = 0xA461
+    CompositeImageExposureTimes = 0xA462
+    Gamma = 0xA500
+    PrintImageMatching = 0xC4A5
+    DNGVersion = 0xC612
+    DNGBackwardVersion = 0xC613
+    UniqueCameraModel = 0xC614
+    LocalizedCameraModel = 0xC615
+    CFAPlaneColor = 0xC616
+    CFALayout = 0xC617
+    LinearizationTable = 0xC618
+    BlackLevelRepeatDim = 0xC619
+    BlackLevel = 0xC61A
+    BlackLevelDeltaH = 0xC61B
+    BlackLevelDeltaV = 0xC61C
+    WhiteLevel = 0xC61D
+    DefaultScale = 0xC61E
+    DefaultCropOrigin = 0xC61F
+    DefaultCropSize = 0xC620
+    ColorMatrix1 = 0xC621
+    ColorMatrix2 = 0xC622
+    CameraCalibration1 = 0xC623
+    CameraCalibration2 = 0xC624
+    ReductionMatrix1 = 0xC625
+    ReductionMatrix2 = 0xC626
+    AnalogBalance = 0xC627
+    AsShotNeutral = 0xC628
+    AsShotWhiteXY = 0xC629
+    BaselineExposure = 0xC62A
+    BaselineNoise = 0xC62B
+    BaselineSharpness = 0xC62C
+    BayerGreenSplit = 0xC62D
+    LinearResponseLimit = 0xC62E
+    CameraSerialNumber = 0xC62F
+    LensInfo = 0xC630
+    ChromaBlurRadius = 0xC631
+    AntiAliasStrength = 0xC632
+    ShadowScale = 0xC633
+    DNGPrivateData = 0xC634
+    MakerNoteSafety = 0xC635
+    CalibrationIlluminant1 = 0xC65A
+    CalibrationIlluminant2 = 0xC65B
+    BestQualityScale = 0xC65C
+    RawDataUniqueID = 0xC65D
+    OriginalRawFileName = 0xC68B
+    OriginalRawFileData = 0xC68C
+    ActiveArea = 0xC68D
+    MaskedAreas = 0xC68E
+    AsShotICCProfile = 0xC68F
+    AsShotPreProfileMatrix = 0xC690
+    CurrentICCProfile = 0xC691
+    CurrentPreProfileMatrix = 0xC692
+    ColorimetricReference = 0xC6BF
+    CameraCalibrationSignature = 0xC6F3
+    ProfileCalibrationSignature = 0xC6F4
+    AsShotProfileName = 0xC6F6
+    NoiseReductionApplied = 0xC6F7
+    ProfileName = 0xC6F8
+    ProfileHueSatMapDims = 0xC6F9
+    ProfileHueSatMapData1 = 0xC6FA
+    ProfileHueSatMapData2 = 0xC6FB
+    ProfileToneCurve = 0xC6FC
+    ProfileEmbedPolicy = 0xC6FD
+    ProfileCopyright = 0xC6FE
+    ForwardMatrix1 = 0xC714
+    ForwardMatrix2 = 0xC715
+    PreviewApplicationName = 0xC716
+    PreviewApplicationVersion = 0xC717
+    PreviewSettingsName = 0xC718
+    PreviewSettingsDigest = 0xC719
+    PreviewColorSpace = 0xC71A
+    PreviewDateTime = 0xC71B
+    RawImageDigest = 0xC71C
+    OriginalRawFileDigest = 0xC71D
+    SubTileBlockSize = 0xC71E
+    RowInterleaveFactor = 0xC71F
+    ProfileLookTableDims = 0xC725
+    ProfileLookTableData = 0xC726
+    OpcodeList1 = 0xC740
+    OpcodeList2 = 0xC741
+    OpcodeList3 = 0xC74E
+    NoiseProfile = 0xC761
+
+
+"""Maps EXIF tags to tag names."""
+TAGS = {
+    **{i.value: i.name for i in Base},
+    0x920C: "SpatialFrequencyResponse",
+    0x9214: "SubjectLocation",
+    0x9215: "ExposureIndex",
+    0x828E: "CFAPattern",
+    0x920B: "FlashEnergy",
+    0x9216: "TIFF/EPStandardID",
+}
+
+
+class GPS(IntEnum):
+    GPSVersionID = 0x00
+    GPSLatitudeRef = 0x01
+    GPSLatitude = 0x02
+    GPSLongitudeRef = 0x03
+    GPSLongitude = 0x04
+    GPSAltitudeRef = 0x05
+    GPSAltitude = 0x06
+    GPSTimeStamp = 0x07
+    GPSSatellites = 0x08
+    GPSStatus = 0x09
+    GPSMeasureMode = 0x0A
+    GPSDOP = 0x0B
+    GPSSpeedRef = 0x0C
+    GPSSpeed = 0x0D
+    GPSTrackRef = 0x0E
+    GPSTrack = 0x0F
+    GPSImgDirectionRef = 0x10
+    GPSImgDirection = 0x11
+    GPSMapDatum = 0x12
+    GPSDestLatitudeRef = 0x13
+    GPSDestLatitude = 0x14
+    GPSDestLongitudeRef = 0x15
+    GPSDestLongitude = 0x16
+    GPSDestBearingRef = 0x17
+    GPSDestBearing = 0x18
+    GPSDestDistanceRef = 0x19
+    GPSDestDistance = 0x1A
+    GPSProcessingMethod = 0x1B
+    GPSAreaInformation = 0x1C
+    GPSDateStamp = 0x1D
+    GPSDifferential = 0x1E
+    GPSHPositioningError = 0x1F
+
+
+"""Maps EXIF GPS tags to tag names."""
+GPSTAGS = {i.value: i.name for i in GPS}
+
+
+class Interop(IntEnum):
+    InteropIndex = 0x0001
+    InteropVersion = 0x0002
+    RelatedImageFileFormat = 0x1000
+    RelatedImageWidth = 0x1001
+    RelatedImageHeight = 0x1002
+
+
+class IFD(IntEnum):
+    Exif = 0x8769
+    GPSInfo = 0x8825
+    MakerNote = 0x927C
+    Makernote = 0x927C  # Deprecated
+    Interop = 0xA005
+    IFD1 = -1
+
+
+class LightSource(IntEnum):
+    Unknown = 0x00
+    Daylight = 0x01
+    Fluorescent = 0x02
+    Tungsten = 0x03
+    Flash = 0x04
+    Fine = 0x09
+    Cloudy = 0x0A
+    Shade = 0x0B
+    DaylightFluorescent = 0x0C
+    DayWhiteFluorescent = 0x0D
+    CoolWhiteFluorescent = 0x0E
+    WhiteFluorescent = 0x0F
+    StandardLightA = 0x11
+    StandardLightB = 0x12
+    StandardLightC = 0x13
+    D55 = 0x14
+    D65 = 0x15
+    D75 = 0x16
+    D50 = 0x17
+    ISO = 0x18
+    Other = 0xFF

venv/Lib/site-packages/PIL/FitsImagePlugin.py

@@ -0,0 +1,152 @@
+#
+# The Python Imaging Library
+# $Id$
+#
+# FITS file handling
+#
+# Copyright (c) 1998-2003 by Fredrik Lundh
+#
+# See the README file for information on usage and redistribution.
+#
+from __future__ import annotations
+
+import gzip
+import math
+
+from . import Image, ImageFile
+
+
+def _accept(prefix: bytes) -> bool:
+    return prefix[:6] == b"SIMPLE"
+
+
+class FitsImageFile(ImageFile.ImageFile):
+    format = "FITS"
+    format_description = "FITS"
+
+    def _open(self) -> None:
+        assert self.fp is not None
+
+        headers: dict[bytes, bytes] = {}
+        header_in_progress = False
+        decoder_name = ""
+        while True:
+            header = self.fp.read(80)
+            if not header:
+                msg = "Truncated FITS file"
+                raise OSError(msg)
+            keyword = header[:8].strip()
+            if keyword in (b"SIMPLE", b"XTENSION"):
+                header_in_progress = True
+            elif headers and not header_in_progress:
+                # This is now a data unit
+                break
+            elif keyword == b"END":
+                # Seek to the end of the header unit
+                self.fp.seek(math.ceil(self.fp.tell() / 2880) * 2880)
+                if not decoder_name:
+                    decoder_name, offset, args = self._parse_headers(headers)
+
+                header_in_progress = False
+                continue
+
+            if decoder_name:
+                # Keep going to read past the headers
+                continue
+
+            value = header[8:].split(b"/")[0].strip()
+            if value.startswith(b"="):
+                value = value[1:].strip()
+            if not headers and (not _accept(keyword) or value != b"T"):
+                msg = "Not a FITS file"
+                raise SyntaxError(msg)
+            headers[keyword] = value
+
+        if not decoder_name:
+            msg = "No image data"
+            raise ValueError(msg)
+
+        offset += self.fp.tell() - 80
+        self.tile = [ImageFile._Tile(decoder_name, (0, 0) + self.size, offset, args)]
+
+    def _get_size(
+        self, headers: dict[bytes, bytes], prefix: bytes
+    ) -> tuple[int, int] | None:
+        naxis = int(headers[prefix + b"NAXIS"])
+        if naxis == 0:
+            return None
+
+        if naxis == 1:
+            return 1, int(headers[prefix + b"NAXIS1"])
+        else:
+            return int(headers[prefix + b"NAXIS1"]), int(headers[prefix + b"NAXIS2"])
+
+    def _parse_headers(
+        self, headers: dict[bytes, bytes]
+    ) -> tuple[str, int, tuple[str | int, ...]]:
+        prefix = b""
+        decoder_name = "raw"
+        offset = 0
+        if (
+            headers.get(b"XTENSION") == b"'BINTABLE'"
+            and headers.get(b"ZIMAGE") == b"T"
+            and headers[b"ZCMPTYPE"] == b"'GZIP_1  '"
+        ):
+            no_prefix_size = self._get_size(headers, prefix) or (0, 0)
+            number_of_bits = int(headers[b"BITPIX"])
+            offset = no_prefix_size[0] * no_prefix_size[1] * (number_of_bits // 8)
+
+            prefix = b"Z"
+            decoder_name = "fits_gzip"
+
+        size = self._get_size(headers, prefix)
+        if not size:
+            return "", 0, ()
+
+        self._size = size
+
+        number_of_bits = int(headers[prefix + b"BITPIX"])
+        if number_of_bits == 8:
+            self._mode = "L"
+        elif number_of_bits == 16:
+            self._mode = "I;16"
+        elif number_of_bits == 32:
+            self._mode = "I"
+        elif number_of_bits in (-32, -64):
+            self._mode = "F"
+
+        args: tuple[str | int, ...]
+        if decoder_name == "raw":
+            args = (self.mode, 0, -1)
+        else:
+            args = (number_of_bits,)
+        return decoder_name, offset, args
+
+
+class FitsGzipDecoder(ImageFile.PyDecoder):
+    _pulls_fd = True
+
+    def decode(self, buffer: bytes | Image.SupportsArrayInterface) -> tuple[int, int]:
+        assert self.fd is not None
+        value = gzip.decompress(self.fd.read())
+
+        rows = []
+        offset = 0
+        number_of_bits = min(self.args[0] // 8, 4)
+        for y in range(self.state.ysize):
+            row = bytearray()
+            for x in range(self.state.xsize):
+                row += value[offset + (4 - number_of_bits) : offset + 4]
+                offset += 4
+            rows.append(row)
+        self.set_as_raw(bytes([pixel for row in rows[::-1] for pixel in row]))
+        return -1, 0
+
+
+# --------------------------------------------------------------------
+# Registry
+
+Image.register_open(FitsImageFile.format, FitsImageFile, _accept)
+Image.register_decoder("fits_gzip", FitsGzipDecoder)
+
+Image.register_extensions(FitsImageFile.format, [".fit", ".fits"])

venv/Lib/site-packages/PIL/FliImagePlugin.py

@@ -0,0 +1,175 @@
+#
+# The Python Imaging Library.
+# $Id$
+#
+# FLI/FLC file handling.
+#
+# History:
+#       95-09-01 fl     Created
+#       97-01-03 fl     Fixed parser, setup decoder tile
+#       98-07-15 fl     Renamed offset attribute to avoid name clash
+#
+# Copyright (c) Secret Labs AB 1997-98.
+# Copyright (c) Fredrik Lundh 1995-97.
+#
+# See the README file for information on usage and redistribution.
+#
+from __future__ import annotations
+
+import os
+
+from . import Image, ImageFile, ImagePalette
+from ._binary import i16le as i16
+from ._binary import i32le as i32
+from ._binary import o8
+
+#
+# decoder
+
+
+def _accept(prefix: bytes) -> bool:
+    return (
+        len(prefix) >= 6
+        and i16(prefix, 4) in [0xAF11, 0xAF12]
+        and i16(prefix, 14) in [0, 3]  # flags
+    )
+
+
+##
+# Image plugin for the FLI/FLC animation format.  Use the <b>seek</b>
+# method to load individual frames.
+
+
+class FliImageFile(ImageFile.ImageFile):
+    format = "FLI"
+    format_description = "Autodesk FLI/FLC Animation"
+    _close_exclusive_fp_after_loading = False
+
+    def _open(self) -> None:
+        # HEAD
+        s = self.fp.read(128)
+        if not (_accept(s) and s[20:22] == b"\x00\x00"):
+            msg = "not an FLI/FLC file"
+            raise SyntaxError(msg)
+
+        # frames
+        self.n_frames = i16(s, 6)
+        self.is_animated = self.n_frames > 1
+
+        # image characteristics
+        self._mode = "P"
+        self._size = i16(s, 8), i16(s, 10)
+
+        # animation speed
+        duration = i32(s, 16)
+        magic = i16(s, 4)
+        if magic == 0xAF11:
+            duration = (duration * 1000) // 70
+        self.info["duration"] = duration
+
+        # look for palette
+        palette = [(a, a, a) for a in range(256)]
+
+        s = self.fp.read(16)
+
+        self.__offset = 128
+
+        if i16(s, 4) == 0xF100:
+            # prefix chunk; ignore it
+            self.__offset = self.__offset + i32(s)
+            self.fp.seek(self.__offset)
+            s = self.fp.read(16)
+
+        if i16(s, 4) == 0xF1FA:
+            # look for palette chunk
+            number_of_subchunks = i16(s, 6)
+            chunk_size: int | None = None
+            for _ in range(number_of_subchunks):
+                if chunk_size is not None:
+                    self.fp.seek(chunk_size - 6, os.SEEK_CUR)
+                s = self.fp.read(6)
+                chunk_type = i16(s, 4)
+                if chunk_type in (4, 11):
+                    self._palette(palette, 2 if chunk_type == 11 else 0)
+                    break
+                chunk_size = i32(s)
+                if not chunk_size:
+                    break
+
+        self.palette = ImagePalette.raw(
+            "RGB", b"".join(o8(r) + o8(g) + o8(b) for (r, g, b) in palette)
+        )
+
+        # set things up to decode first frame
+        self.__frame = -1
+        self._fp = self.fp
+        self.__rewind = self.fp.tell()
+        self.seek(0)
+
+    def _palette(self, palette: list[tuple[int, int, int]], shift: int) -> None:
+        # load palette
+
+        i = 0
+        for e in range(i16(self.fp.read(2))):
+            s = self.fp.read(2)
+            i = i + s[0]
+            n = s[1]
+            if n == 0:
+                n = 256
+            s = self.fp.read(n * 3)
+            for n in range(0, len(s), 3):
+                r = s[n] << shift
+                g = s[n + 1] << shift
+                b = s[n + 2] << shift
+                palette[i] = (r, g, b)
+                i += 1
+
+    def seek(self, frame: int) -> None:
+        if not self._seek_check(frame):
+            return
+        if frame < self.__frame:
+            self._seek(0)
+
+        for f in range(self.__frame + 1, frame + 1):
+            self._seek(f)
+
+    def _seek(self, frame: int) -> None:
+        if frame == 0:
+            self.__frame = -1
+            self._fp.seek(self.__rewind)
+            self.__offset = 128
+        else:
+            # ensure that the previous frame was loaded
+            self.load()
+
+        if frame != self.__frame + 1:
+            msg = f"cannot seek to frame {frame}"
+            raise ValueError(msg)
+        self.__frame = frame
+
+        # move to next frame
+        self.fp = self._fp
+        self.fp.seek(self.__offset)
+
+        s = self.fp.read(4)
+        if not s:
+            msg = "missing frame size"
+            raise EOFError(msg)
+
+        framesize = i32(s)
+
+        self.decodermaxblock = framesize
+        self.tile = [ImageFile._Tile("fli", (0, 0) + self.size, self.__offset)]
+
+        self.__offset += framesize
+
+    def tell(self) -> int:
+        return self.__frame
+
+
+#
+# registry
+
+Image.register_open(FliImageFile.format, FliImageFile, _accept)
+
+Image.register_extensions(FliImageFile.format, [".fli", ".flc"])

venv/Lib/site-packages/PIL/FontFile.py

@@ -0,0 +1,134 @@
+#
+# The Python Imaging Library
+# $Id$
+#
+# base class for raster font file parsers
+#
+# history:
+# 1997-06-05 fl   created
+# 1997-08-19 fl   restrict image width
+#
+# Copyright (c) 1997-1998 by Secret Labs AB
+# Copyright (c) 1997-1998 by Fredrik Lundh
+#
+# See the README file for information on usage and redistribution.
+#
+from __future__ import annotations
+
+import os
+from typing import BinaryIO
+
+from . import Image, _binary
+
+WIDTH = 800
+
+
+def puti16(
+    fp: BinaryIO, values: tuple[int, int, int, int, int, int, int, int, int, int]
+) -> None:
+    """Write network order (big-endian) 16-bit sequence"""
+    for v in values:
+        if v < 0:
+            v += 65536
+        fp.write(_binary.o16be(v))
+
+
+class FontFile:
+    """Base class for raster font file handlers."""
+
+    bitmap: Image.Image | None = None
+
+    def __init__(self) -> None:
+        self.info: dict[bytes, bytes | int] = {}
+        self.glyph: list[
+            tuple[
+                tuple[int, int],
+                tuple[int, int, int, int],
+                tuple[int, int, int, int],
+                Image.Image,
+            ]
+            | None
+        ] = [None] * 256
+
+    def __getitem__(self, ix: int) -> (
+        tuple[
+            tuple[int, int],
+            tuple[int, int, int, int],
+            tuple[int, int, int, int],
+            Image.Image,
+        ]
+        | None
+    ):
+        return self.glyph[ix]
+
+    def compile(self) -> None:
+        """Create metrics and bitmap"""
+
+        if self.bitmap:
+            return
+
+        # create bitmap large enough to hold all data
+        h = w = maxwidth = 0
+        lines = 1
+        for glyph in self.glyph:
+            if glyph:
+                d, dst, src, im = glyph
+                h = max(h, src[3] - src[1])
+                w = w + (src[2] - src[0])
+                if w > WIDTH:
+                    lines += 1
+                    w = src[2] - src[0]
+                maxwidth = max(maxwidth, w)
+
+        xsize = maxwidth
+        ysize = lines * h
+
+        if xsize == 0 and ysize == 0:
+            return
+
+        self.ysize = h
+
+        # paste glyphs into bitmap
+        self.bitmap = Image.new("1", (xsize, ysize))
+        self.metrics: list[
+            tuple[tuple[int, int], tuple[int, int, int, int], tuple[int, int, int, int]]
+            | None
+        ] = [None] * 256
+        x = y = 0
+        for i in range(256):
+            glyph = self[i]
+            if glyph:
+                d, dst, src, im = glyph
+                xx = src[2] - src[0]
+                x0, y0 = x, y
+                x = x + xx
+                if x > WIDTH:
+                    x, y = 0, y + h
+                    x0, y0 = x, y
+                    x = xx
+                s = src[0] + x0, src[1] + y0, src[2] + x0, src[3] + y0
+                self.bitmap.paste(im.crop(src), s)
+                self.metrics[i] = d, dst, s
+
+    def save(self, filename: str) -> None:
+        """Save font"""
+
+        self.compile()
+
+        # font data
+        if not self.bitmap:
+            msg = "No bitmap created"
+            raise ValueError(msg)
+        self.bitmap.save(os.path.splitext(filename)[0] + ".pbm", "PNG")
+
+        # font metrics
+        with open(os.path.splitext(filename)[0] + ".pil", "wb") as fp:
+            fp.write(b"PILfont\n")
+            fp.write(f";;;;;;{self.ysize};\n".encode("ascii"))  # HACK!!!
+            fp.write(b"DATA\n")
+            for id in range(256):
+                m = self.metrics[id]
+                if not m:
+                    puti16(fp, (0,) * 10)
+                else:
+                    puti16(fp, m[0] + m[1] + m[2])

venv/Lib/site-packages/PIL/FpxImagePlugin.py

@@ -0,0 +1,257 @@
+#
+# THIS IS WORK IN PROGRESS
+#
+# The Python Imaging Library.
+# $Id$
+#
+# FlashPix support for PIL
+#
+# History:
+# 97-01-25 fl   Created (reads uncompressed RGB images only)
+#
+# Copyright (c) Secret Labs AB 1997.
+# Copyright (c) Fredrik Lundh 1997.
+#
+# See the README file for information on usage and redistribution.
+#
+from __future__ import annotations
+
+import olefile
+
+from . import Image, ImageFile
+from ._binary import i32le as i32
+
+# we map from colour field tuples to (mode, rawmode) descriptors
+MODES = {
+    # opacity
+    (0x00007FFE,): ("A", "L"),
+    # monochrome
+    (0x00010000,): ("L", "L"),
+    (0x00018000, 0x00017FFE): ("RGBA", "LA"),
+    # photo YCC
+    (0x00020000, 0x00020001, 0x00020002): ("RGB", "YCC;P"),
+    (0x00028000, 0x00028001, 0x00028002, 0x00027FFE): ("RGBA", "YCCA;P"),
+    # standard RGB (NIFRGB)
+    (0x00030000, 0x00030001, 0x00030002): ("RGB", "RGB"),
+    (0x00038000, 0x00038001, 0x00038002, 0x00037FFE): ("RGBA", "RGBA"),
+}
+
+
+#
+# --------------------------------------------------------------------
+
+
+def _accept(prefix: bytes) -> bool:
+    return prefix[:8] == olefile.MAGIC
+
+
+##
+# Image plugin for the FlashPix images.
+
+
+class FpxImageFile(ImageFile.ImageFile):
+    format = "FPX"
+    format_description = "FlashPix"
+
+    def _open(self) -> None:
+        #
+        # read the OLE directory and see if this is a likely
+        # to be a FlashPix file
+
+        try:
+            self.ole = olefile.OleFileIO(self.fp)
+        except OSError as e:
+            msg = "not an FPX file; invalid OLE file"
+            raise SyntaxError(msg) from e
+
+        root = self.ole.root
+        if not root or root.clsid != "56616700-C154-11CE-8553-00AA00A1F95B":
+            msg = "not an FPX file; bad root CLSID"
+            raise SyntaxError(msg)
+
+        self._open_index(1)
+
+    def _open_index(self, index: int = 1) -> None:
+        #
+        # get the Image Contents Property Set
+
+        prop = self.ole.getproperties(
+            [f"Data Object Store {index:06d}", "\005Image Contents"]
+        )
+
+        # size (highest resolution)
+
+        assert isinstance(prop[0x1000002], int)
+        assert isinstance(prop[0x1000003], int)
+        self._size = prop[0x1000002], prop[0x1000003]
+
+        size = max(self.size)
+        i = 1
+        while size > 64:
+            size = size // 2
+            i += 1
+        self.maxid = i - 1
+
+        # mode.  instead of using a single field for this, flashpix
+        # requires you to specify the mode for each channel in each
+        # resolution subimage, and leaves it to the decoder to make
+        # sure that they all match.  for now, we'll cheat and assume
+        # that this is always the case.
+
+        id = self.maxid << 16
+
+        s = prop[0x2000002 | id]
+
+        if not isinstance(s, bytes) or (bands := i32(s, 4)) > 4:
+            msg = "Invalid number of bands"
+            raise OSError(msg)
+
+        # note: for now, we ignore the "uncalibrated" flag
+        colors = tuple(i32(s, 8 + i * 4) & 0x7FFFFFFF for i in range(bands))
+
+        self._mode, self.rawmode = MODES[colors]
+
+        # load JPEG tables, if any
+        self.jpeg = {}
+        for i in range(256):
+            id = 0x3000001 | (i << 16)
+            if id in prop:
+                self.jpeg[i] = prop[id]
+
+        self._open_subimage(1, self.maxid)
+
+    def _open_subimage(self, index: int = 1, subimage: int = 0) -> None:
+        #
+        # setup tile descriptors for a given subimage
+
+        stream = [
+            f"Data Object Store {index:06d}",
+            f"Resolution {subimage:04d}",
+            "Subimage 0000 Header",
+        ]
+
+        fp = self.ole.openstream(stream)
+
+        # skip prefix
+        fp.read(28)
+
+        # header stream
+        s = fp.read(36)
+
+        size = i32(s, 4), i32(s, 8)
+        # tilecount = i32(s, 12)
+        tilesize = i32(s, 16), i32(s, 20)
+        # channels = i32(s, 24)
+        offset = i32(s, 28)
+        length = i32(s, 32)
+
+        if size != self.size:
+            msg = "subimage mismatch"
+            raise OSError(msg)
+
+        # get tile descriptors
+        fp.seek(28 + offset)
+        s = fp.read(i32(s, 12) * length)
+
+        x = y = 0
+        xsize, ysize = size
+        xtile, ytile = tilesize
+        self.tile = []
+
+        for i in range(0, len(s), length):
+            x1 = min(xsize, x + xtile)
+            y1 = min(ysize, y + ytile)
+
+            compression = i32(s, i + 8)
+
+            if compression == 0:
+                self.tile.append(
+                    ImageFile._Tile(
+                        "raw",
+                        (x, y, x1, y1),
+                        i32(s, i) + 28,
+                        self.rawmode,
+                    )
+                )
+
+            elif compression == 1:
+                # FIXME: the fill decoder is not implemented
+                self.tile.append(
+                    ImageFile._Tile(
+                        "fill",
+                        (x, y, x1, y1),
+                        i32(s, i) + 28,
+                        (self.rawmode, s[12:16]),
+                    )
+                )
+
+            elif compression == 2:
+                internal_color_conversion = s[14]
+                jpeg_tables = s[15]
+                rawmode = self.rawmode
+
+                if internal_color_conversion:
+                    # The image is stored as usual (usually YCbCr).
+                    if rawmode == "RGBA":
+                        # For "RGBA", data is stored as YCbCrA based on
+                        # negative RGB. The following trick works around
+                        # this problem :
+                        jpegmode, rawmode = "YCbCrK", "CMYK"
+                    else:
+                        jpegmode = None  # let the decoder decide
+
+                else:
+                    # The image is stored as defined by rawmode
+                    jpegmode = rawmode
+
+                self.tile.append(
+                    ImageFile._Tile(
+                        "jpeg",
+                        (x, y, x1, y1),
+                        i32(s, i) + 28,
+                        (rawmode, jpegmode),
+                    )
+                )
+
+                # FIXME: jpeg tables are tile dependent; the prefix
+                # data must be placed in the tile descriptor itself!
+
+                if jpeg_tables:
+                    self.tile_prefix = self.jpeg[jpeg_tables]
+
+            else:
+                msg = "unknown/invalid compression"
+                raise OSError(msg)
+
+            x = x + xtile
+            if x >= xsize:
+                x, y = 0, y + ytile
+                if y >= ysize:
+                    break  # isn't really required
+
+        self.stream = stream
+        self._fp = self.fp
+        self.fp = None
+
+    def load(self) -> Image.core.PixelAccess | None:
+        if not self.fp:
+            self.fp = self.ole.openstream(self.stream[:2] + ["Subimage 0000 Data"])
+
+        return ImageFile.ImageFile.load(self)
+
+    def close(self) -> None:
+        self.ole.close()
+        super().close()
+
+    def __exit__(self, *args: object) -> None:
+        self.ole.close()
+        super().__exit__()
+
+
+#
+# --------------------------------------------------------------------
+
+
+Image.register_open(FpxImageFile.format, FpxImageFile, _accept)
+
+Image.register_extension(FpxImageFile.format, ".fpx")

venv/Lib/site-packages/PIL/FtexImagePlugin.py

@@ -0,0 +1,115 @@
+"""
+A Pillow loader for .ftc and .ftu files (FTEX)
+Jerome Leclanche <jerome@leclan.ch>
+
+The contents of this file are hereby released in the public domain (CC0)
+Full text of the CC0 license:
+  https://creativecommons.org/publicdomain/zero/1.0/
+
+Independence War 2: Edge Of Chaos - Texture File Format - 16 October 2001
+
+The textures used for 3D objects in Independence War 2: Edge Of Chaos are in a
+packed custom format called FTEX. This file format uses file extensions FTC
+and FTU.
+* FTC files are compressed textures (using standard texture compression).
+* FTU files are not compressed.
+Texture File Format
+The FTC and FTU texture files both use the same format. This
+has the following structure:
+{header}
+{format_directory}
+{data}
+Where:
+{header} = {
+    u32:magic,
+    u32:version,
+    u32:width,
+    u32:height,
+    u32:mipmap_count,
+    u32:format_count
+}
+
+* The "magic" number is "FTEX".
+* "width" and "height" are the dimensions of the texture.
+* "mipmap_count" is the number of mipmaps in the texture.
+* "format_count" is the number of texture formats (different versions of the
+same texture) in this file.
+
+{format_directory} = format_count * { u32:format, u32:where }
+
+The format value is 0 for DXT1 compressed textures and 1 for 24-bit RGB
+uncompressed textures.
+The texture data for a format starts at the position "where" in the file.
+
+Each set of texture data in the file has the following structure:
+{data} = format_count * { u32:mipmap_size, mipmap_size * { u8 } }
+* "mipmap_size" is the number of bytes in that mip level. For compressed
+textures this is the size of the texture data compressed with DXT1. For 24 bit
+uncompressed textures, this is 3 * width * height. Following this are the image
+bytes for that mipmap level.
+
+Note: All data is stored in little-Endian (Intel) byte order.
+"""
+
+from __future__ import annotations
+
+import struct
+from enum import IntEnum
+from io import BytesIO
+
+from . import Image, ImageFile
+
+MAGIC = b"FTEX"
+
+
+class Format(IntEnum):
+    DXT1 = 0
+    UNCOMPRESSED = 1
+
+
+class FtexImageFile(ImageFile.ImageFile):
+    format = "FTEX"
+    format_description = "Texture File Format (IW2:EOC)"
+
+    def _open(self) -> None:
+        if not _accept(self.fp.read(4)):
+            msg = "not an FTEX file"
+            raise SyntaxError(msg)
+        struct.unpack("<i", self.fp.read(4))  # version
+        self._size = struct.unpack("<2i", self.fp.read(8))
+        mipmap_count, format_count = struct.unpack("<2i", self.fp.read(8))
+
+        self._mode = "RGB"
+
+        # Only support single-format files.
+        # I don't know of any multi-format file.
+        assert format_count == 1
+
+        format, where = struct.unpack("<2i", self.fp.read(8))
+        self.fp.seek(where)
+        (mipmap_size,) = struct.unpack("<i", self.fp.read(4))
+
+        data = self.fp.read(mipmap_size)
+
+        if format == Format.DXT1:
+            self._mode = "RGBA"
+            self.tile = [ImageFile._Tile("bcn", (0, 0) + self.size, 0, (1,))]
+        elif format == Format.UNCOMPRESSED:
+            self.tile = [ImageFile._Tile("raw", (0, 0) + self.size, 0, "RGB")]
+        else:
+            msg = f"Invalid texture compression format: {repr(format)}"
+            raise ValueError(msg)
+
+        self.fp.close()
+        self.fp = BytesIO(data)
+
+    def load_seek(self, pos: int) -> None:
+        pass
+
+
+def _accept(prefix: bytes) -> bool:
+    return prefix[:4] == MAGIC
+
+
+Image.register_open(FtexImageFile.format, FtexImageFile, _accept)
+Image.register_extensions(FtexImageFile.format, [".ftc", ".ftu"])

venv/Lib/site-packages/PIL/GbrImagePlugin.py

@@ -0,0 +1,103 @@
+#
+# The Python Imaging Library
+#
+# load a GIMP brush file
+#
+# History:
+#       96-03-14 fl     Created
+#       16-01-08 es     Version 2
+#
+# Copyright (c) Secret Labs AB 1997.
+# Copyright (c) Fredrik Lundh 1996.
+# Copyright (c) Eric Soroos 2016.
+#
+# See the README file for information on usage and redistribution.
+#
+#
+# See https://github.com/GNOME/gimp/blob/mainline/devel-docs/gbr.txt for
+# format documentation.
+#
+# This code Interprets version 1 and 2 .gbr files.
+# Version 1 files are obsolete, and should not be used for new
+#   brushes.
+# Version 2 files are saved by GIMP v2.8 (at least)
+# Version 3 files have a format specifier of 18 for 16bit floats in
+#   the color depth field. This is currently unsupported by Pillow.
+from __future__ import annotations
+
+from . import Image, ImageFile
+from ._binary import i32be as i32
+
+
+def _accept(prefix: bytes) -> bool:
+    return len(prefix) >= 8 and i32(prefix, 0) >= 20 and i32(prefix, 4) in (1, 2)
+
+
+##
+# Image plugin for the GIMP brush format.
+
+
+class GbrImageFile(ImageFile.ImageFile):
+    format = "GBR"
+    format_description = "GIMP brush file"
+
+    def _open(self) -> None:
+        header_size = i32(self.fp.read(4))
+        if header_size < 20:
+            msg = "not a GIMP brush"
+            raise SyntaxError(msg)
+        version = i32(self.fp.read(4))
+        if version not in (1, 2):
+            msg = f"Unsupported GIMP brush version: {version}"
+            raise SyntaxError(msg)
+
+        width = i32(self.fp.read(4))
+        height = i32(self.fp.read(4))
+        color_depth = i32(self.fp.read(4))
+        if width <= 0 or height <= 0:
+            msg = "not a GIMP brush"
+            raise SyntaxError(msg)
+        if color_depth not in (1, 4):
+            msg = f"Unsupported GIMP brush color depth: {color_depth}"
+            raise SyntaxError(msg)
+
+        if version == 1:
+            comment_length = header_size - 20
+        else:
+            comment_length = header_size - 28
+            magic_number = self.fp.read(4)
+            if magic_number != b"GIMP":
+                msg = "not a GIMP brush, bad magic number"
+                raise SyntaxError(msg)
+            self.info["spacing"] = i32(self.fp.read(4))
+
+        comment = self.fp.read(comment_length)[:-1]
+
+        if color_depth == 1:
+            self._mode = "L"
+        else:
+            self._mode = "RGBA"
+
+        self._size = width, height
+
+        self.info["comment"] = comment
+
+        # Image might not be small
+        Image._decompression_bomb_check(self.size)
+
+        # Data is an uncompressed block of w * h * bytes/pixel
+        self._data_size = width * height * color_depth
+
+    def load(self) -> Image.core.PixelAccess | None:
+        if self._im is None:
+            self.im = Image.core.new(self.mode, self.size)
+            self.frombytes(self.fp.read(self._data_size))
+        return Image.Image.load(self)
+
+
+#
+# registry
+
+
+Image.register_open(GbrImageFile.format, GbrImageFile, _accept)
+Image.register_extension(GbrImageFile.format, ".gbr")

venv/Lib/site-packages/PIL/GdImageFile.py

@@ -0,0 +1,102 @@
+#
+# The Python Imaging Library.
+# $Id$
+#
+# GD file handling
+#
+# History:
+# 1996-04-12 fl   Created
+#
+# Copyright (c) 1997 by Secret Labs AB.
+# Copyright (c) 1996 by Fredrik Lundh.
+#
+# See the README file for information on usage and redistribution.
+#
+
+
+"""
+.. note::
+    This format cannot be automatically recognized, so the
+    class is not registered for use with :py:func:`PIL.Image.open()`.  To open a
+    gd file, use the :py:func:`PIL.GdImageFile.open()` function instead.
+
+.. warning::
+    THE GD FORMAT IS NOT DESIGNED FOR DATA INTERCHANGE.  This
+    implementation is provided for convenience and demonstrational
+    purposes only.
+"""
+from __future__ import annotations
+
+from typing import IO
+
+from . import ImageFile, ImagePalette, UnidentifiedImageError
+from ._binary import i16be as i16
+from ._binary import i32be as i32
+from ._typing import StrOrBytesPath
+
+
+class GdImageFile(ImageFile.ImageFile):
+    """
+    Image plugin for the GD uncompressed format.  Note that this format
+    is not supported by the standard :py:func:`PIL.Image.open()` function.  To use
+    this plugin, you have to import the :py:mod:`PIL.GdImageFile` module and
+    use the :py:func:`PIL.GdImageFile.open()` function.
+    """
+
+    format = "GD"
+    format_description = "GD uncompressed images"
+
+    def _open(self) -> None:
+        # Header
+        assert self.fp is not None
+
+        s = self.fp.read(1037)
+
+        if i16(s) not in [65534, 65535]:
+            msg = "Not a valid GD 2.x .gd file"
+            raise SyntaxError(msg)
+
+        self._mode = "L"  # FIXME: "P"
+        self._size = i16(s, 2), i16(s, 4)
+
+        true_color = s[6]
+        true_color_offset = 2 if true_color else 0
+
+        # transparency index
+        tindex = i32(s, 7 + true_color_offset)
+        if tindex < 256:
+            self.info["transparency"] = tindex
+
+        self.palette = ImagePalette.raw(
+            "XBGR", s[7 + true_color_offset + 4 : 7 + true_color_offset + 4 + 256 * 4]
+        )
+
+        self.tile = [
+            ImageFile._Tile(
+                "raw",
+                (0, 0) + self.size,
+                7 + true_color_offset + 4 + 256 * 4,
+                "L",
+            )
+        ]
+
+
+def open(fp: StrOrBytesPath | IO[bytes], mode: str = "r") -> GdImageFile:
+    """
+    Load texture from a GD image file.
+
+    :param fp: GD file name, or an opened file handle.
+    :param mode: Optional mode.  In this version, if the mode argument
+        is given, it must be "r".
+    :returns: An image instance.
+    :raises OSError: If the image could not be read.
+    """
+    if mode != "r":
+        msg = "bad mode"
+        raise ValueError(msg)
+
+    try:
+        return GdImageFile(fp)
+    except SyntaxError as e:
+        msg = "cannot identify this image file"
+        raise UnidentifiedImageError(msg) from e

venv/Lib/site-packages/PIL/GimpGradientFile.py

@@ -0,0 +1,149 @@
+#
+# Python Imaging Library
+# $Id$
+#
+# stuff to read (and render) GIMP gradient files
+#
+# History:
+#       97-08-23 fl     Created
+#
+# Copyright (c) Secret Labs AB 1997.
+# Copyright (c) Fredrik Lundh 1997.
+#
+# See the README file for information on usage and redistribution.
+#
+
+"""
+Stuff to translate curve segments to palette values (derived from
+the corresponding code in GIMP, written by Federico Mena Quintero.
+See the GIMP distribution for more information.)
+"""
+from __future__ import annotations
+
+from math import log, pi, sin, sqrt
+from typing import IO, Callable
+
+from ._binary import o8
+
+EPSILON = 1e-10
+""""""  # Enable auto-doc for data member
+
+
+def linear(middle: float, pos: float) -> float:
+    if pos <= middle:
+        if middle < EPSILON:
+            return 0.0
+        else:
+            return 0.5 * pos / middle
+    else:
+        pos = pos - middle
+        middle = 1.0 - middle
+        if middle < EPSILON:
+            return 1.0
+        else:
+            return 0.5 + 0.5 * pos / middle
+
+
+def curved(middle: float, pos: float) -> float:
+    return pos ** (log(0.5) / log(max(middle, EPSILON)))
+
+
+def sine(middle: float, pos: float) -> float:
+    return (sin((-pi / 2.0) + pi * linear(middle, pos)) + 1.0) / 2.0
+
+
+def sphere_increasing(middle: float, pos: float) -> float:
+    return sqrt(1.0 - (linear(middle, pos) - 1.0) ** 2)
+
+
+def sphere_decreasing(middle: float, pos: float) -> float:
+    return 1.0 - sqrt(1.0 - linear(middle, pos) ** 2)
+
+
+SEGMENTS = [linear, curved, sine, sphere_increasing, sphere_decreasing]
+""""""  # Enable auto-doc for data member
+
+
+class GradientFile:
+    gradient: (
+        list[
+            tuple[
+                float,
+                float,
+                float,
+                list[float],
+                list[float],
+                Callable[[float, float], float],
+            ]
+        ]
+        | None
+    ) = None
+
+    def getpalette(self, entries: int = 256) -> tuple[bytes, str]:
+        assert self.gradient is not None
+        palette = []
+
+        ix = 0
+        x0, x1, xm, rgb0, rgb1, segment = self.gradient[ix]
+
+        for i in range(entries):
+            x = i / (entries - 1)
+
+            while x1 < x:
+                ix += 1
+                x0, x1, xm, rgb0, rgb1, segment = self.gradient[ix]
+
+            w = x1 - x0
+
+            if w < EPSILON:
+                scale = segment(0.5, 0.5)
+            else:
+                scale = segment((xm - x0) / w, (x - x0) / w)
+
+            # expand to RGBA
+            r = o8(int(255 * ((rgb1[0] - rgb0[0]) * scale + rgb0[0]) + 0.5))
+            g = o8(int(255 * ((rgb1[1] - rgb0[1]) * scale + rgb0[1]) + 0.5))
+            b = o8(int(255 * ((rgb1[2] - rgb0[2]) * scale + rgb0[2]) + 0.5))
+            a = o8(int(255 * ((rgb1[3] - rgb0[3]) * scale + rgb0[3]) + 0.5))
+
+            # add to palette
+            palette.append(r + g + b + a)
+
+        return b"".join(palette), "RGBA"
+
+
+class GimpGradientFile(GradientFile):
+    """File handler for GIMP's gradient format."""
+
+    def __init__(self, fp: IO[bytes]) -> None:
+        if fp.readline()[:13] != b"GIMP Gradient":
+            msg = "not a GIMP gradient file"
+            raise SyntaxError(msg)
+
+        line = fp.readline()
+
+        # GIMP 1.2 gradient files don't contain a name, but GIMP 1.3 files do
+        if line.startswith(b"Name: "):
+            line = fp.readline().strip()
+
+        count = int(line)
+
+        self.gradient = []
+
+        for i in range(count):
+            s = fp.readline().split()
+            w = [float(x) for x in s[:11]]
+
+            x0, x1 = w[0], w[2]
+            xm = w[1]
+            rgb0 = w[3:7]
+            rgb1 = w[7:11]
+
+            segment = SEGMENTS[int(s[11])]
+            cspace = int(s[12])
+
+            if cspace != 0:
+                msg = "cannot handle HSV colour space"
+                raise OSError(msg)
+
+            self.gradient.append((x0, x1, xm, rgb0, rgb1, segment))

venv/Lib/site-packages/PIL/GimpPaletteFile.py

@@ -0,0 +1,58 @@
+#
+# Python Imaging Library
+# $Id$
+#
+# stuff to read GIMP palette files
+#
+# History:
+# 1997-08-23 fl     Created
+# 2004-09-07 fl     Support GIMP 2.0 palette files.
+#
+# Copyright (c) Secret Labs AB 1997-2004.  All rights reserved.
+# Copyright (c) Fredrik Lundh 1997-2004.
+#
+# See the README file for information on usage and redistribution.
+#
+from __future__ import annotations
+
+import re
+from typing import IO
+
+from ._binary import o8
+
+
+class GimpPaletteFile:
+    """File handler for GIMP's palette format."""
+
+    rawmode = "RGB"
+
+    def __init__(self, fp: IO[bytes]) -> None:
+        palette = [o8(i) * 3 for i in range(256)]
+
+        if fp.readline()[:12] != b"GIMP Palette":
+            msg = "not a GIMP palette file"
+            raise SyntaxError(msg)
+
+        for i in range(256):
+            s = fp.readline()
+            if not s:
+                break
+
+            # skip fields and comment lines
+            if re.match(rb"\w+:|#", s):
+                continue
+            if len(s) > 100:
+                msg = "bad palette file"
+                raise SyntaxError(msg)
+
+            v = tuple(map(int, s.split()[:3]))
+            if len(v) != 3:
+                msg = "bad palette entry"
+                raise ValueError(msg)
+
+            palette[i] = o8(v[0]) + o8(v[1]) + o8(v[2])
+
+        self.palette = b"".join(palette)
+
+    def getpalette(self) -> tuple[bytes, str]:
+        return self.palette, self.rawmode

venv/Lib/site-packages/PIL/GribStubImagePlugin.py

@@ -0,0 +1,76 @@
+#
+# The Python Imaging Library
+# $Id$
+#
+# GRIB stub adapter
+#
+# Copyright (c) 1996-2003 by Fredrik Lundh
+#
+# See the README file for information on usage and redistribution.
+#
+from __future__ import annotations
+
+from typing import IO
+
+from . import Image, ImageFile
+
+_handler = None
+
+
+def register_handler(handler: ImageFile.StubHandler | None) -> None:
+    """
+    Install application-specific GRIB image handler.
+
+    :param handler: Handler object.
+    """
+    global _handler
+    _handler = handler
+
+
+# --------------------------------------------------------------------
+# Image adapter
+
+
+def _accept(prefix: bytes) -> bool:
+    return prefix[:4] == b"GRIB" and prefix[7] == 1
+
+
+class GribStubImageFile(ImageFile.StubImageFile):
+    format = "GRIB"
+    format_description = "GRIB"
+
+    def _open(self) -> None:
+        offset = self.fp.tell()
+
+        if not _accept(self.fp.read(8)):
+            msg = "Not a GRIB file"
+            raise SyntaxError(msg)
+
+        self.fp.seek(offset)
+
+        # make something up
+        self._mode = "F"
+        self._size = 1, 1
+
+        loader = self._load()
+        if loader:
+            loader.open(self)
+
+    def _load(self) -> ImageFile.StubHandler | None:
+        return _handler
+
+
+def _save(im: Image.Image, fp: IO[bytes], filename: str | bytes) -> None:
+    if _handler is None or not hasattr(_handler, "save"):
+        msg = "GRIB save handler not installed"
+        raise OSError(msg)
+    _handler.save(im, fp, filename)
+
+
+# --------------------------------------------------------------------
+# Registry
+
+Image.register_open(GribStubImageFile.format, GribStubImageFile, _accept)
+Image.register_save(GribStubImageFile.format, _save)
+
+Image.register_extension(GribStubImageFile.format, ".grib")

venv/Lib/site-packages/PIL/Hdf5StubImagePlugin.py

@@ -0,0 +1,76 @@
+#
+# The Python Imaging Library
+# $Id$
+#
+# HDF5 stub adapter
+#
+# Copyright (c) 2000-2003 by Fredrik Lundh
+#
+# See the README file for information on usage and redistribution.
+#
+from __future__ import annotations
+
+from typing import IO
+
+from . import Image, ImageFile
+
+_handler = None
+
+
+def register_handler(handler: ImageFile.StubHandler | None) -> None:
+    """
+    Install application-specific HDF5 image handler.
+
+    :param handler: Handler object.
+    """
+    global _handler
+    _handler = handler
+
+
+# --------------------------------------------------------------------
+# Image adapter
+
+
+def _accept(prefix: bytes) -> bool:
+    return prefix[:8] == b"\x89HDF\r\n\x1a\n"
+
+
+class HDF5StubImageFile(ImageFile.StubImageFile):
+    format = "HDF5"
+    format_description = "HDF5"
+
+    def _open(self) -> None:
+        offset = self.fp.tell()
+
+        if not _accept(self.fp.read(8)):
+            msg = "Not an HDF file"
+            raise SyntaxError(msg)
+
+        self.fp.seek(offset)
+
+        # make something up
+        self._mode = "F"
+        self._size = 1, 1
+
+        loader = self._load()
+        if loader:
+            loader.open(self)
+
+    def _load(self) -> ImageFile.StubHandler | None:
+        return _handler
+
+
+def _save(im: Image.Image, fp: IO[bytes], filename: str | bytes) -> None:
+    if _handler is None or not hasattr(_handler, "save"):
+        msg = "HDF5 save handler not installed"
+        raise OSError(msg)
+    _handler.save(im, fp, filename)
+
+
+# --------------------------------------------------------------------
+# Registry
+
+Image.register_open(HDF5StubImageFile.format, HDF5StubImageFile, _accept)
+Image.register_save(HDF5StubImageFile.format, _save)
+
+Image.register_extensions(HDF5StubImageFile.format, [".h5", ".hdf"])

venv/Lib/site-packages/PIL/IcnsImagePlugin.py

@@ -0,0 +1,412 @@
+#
+# The Python Imaging Library.
+# $Id$
+#
+# macOS icns file decoder, based on icns.py by Bob Ippolito.
+#
+# history:
+# 2004-10-09 fl   Turned into a PIL plugin; removed 2.3 dependencies.
+# 2020-04-04      Allow saving on all operating systems.
+#
+# Copyright (c) 2004 by Bob Ippolito.
+# Copyright (c) 2004 by Secret Labs.
+# Copyright (c) 2004 by Fredrik Lundh.
+# Copyright (c) 2014 by Alastair Houghton.
+# Copyright (c) 2020 by Pan Jing.
+#
+# See the README file for information on usage and redistribution.
+#
+from __future__ import annotations
+
+import io
+import os
+import struct
+import sys
+from typing import IO
+
+from . import Image, ImageFile, PngImagePlugin, features
+from ._deprecate import deprecate
+
+enable_jpeg2k = features.check_codec("jpg_2000")
+if enable_jpeg2k:
+    from . import Jpeg2KImagePlugin
+
+MAGIC = b"icns"
+HEADERSIZE = 8
+
+
+def nextheader(fobj: IO[bytes]) -> tuple[bytes, int]:
+    return struct.unpack(">4sI", fobj.read(HEADERSIZE))
+
+
+def read_32t(
+    fobj: IO[bytes], start_length: tuple[int, int], size: tuple[int, int, int]
+) -> dict[str, Image.Image]:
+    # The 128x128 icon seems to have an extra header for some reason.
+    (start, length) = start_length
+    fobj.seek(start)
+    sig = fobj.read(4)
+    if sig != b"\x00\x00\x00\x00":
+        msg = "Unknown signature, expecting 0x00000000"
+        raise SyntaxError(msg)
+    return read_32(fobj, (start + 4, length - 4), size)
+
+
+def read_32(
+    fobj: IO[bytes], start_length: tuple[int, int], size: tuple[int, int, int]
+) -> dict[str, Image.Image]:
+    """
+    Read a 32bit RGB icon resource.  Seems to be either uncompressed or
+    an RLE packbits-like scheme.
+    """
+    (start, length) = start_length
+    fobj.seek(start)
+    pixel_size = (size[0] * size[2], size[1] * size[2])
+    sizesq = pixel_size[0] * pixel_size[1]
+    if length == sizesq * 3:
+        # uncompressed ("RGBRGBGB")
+        indata = fobj.read(length)
+        im = Image.frombuffer("RGB", pixel_size, indata, "raw", "RGB", 0, 1)
+    else:
+        # decode image
+        im = Image.new("RGB", pixel_size, None)
+        for band_ix in range(3):
+            data = []
+            bytesleft = sizesq
+            while bytesleft > 0:
+                byte = fobj.read(1)
+                if not byte:
+                    break
+                byte_int = byte[0]
+                if byte_int & 0x80:
+                    blocksize = byte_int - 125
+                    byte = fobj.read(1)
+                    for i in range(blocksize):
+                        data.append(byte)
+                else:
+                    blocksize = byte_int + 1
+                    data.append(fobj.read(blocksize))
+                bytesleft -= blocksize
+                if bytesleft <= 0:
+                    break
+            if bytesleft != 0:
+                msg = f"Error reading channel [{repr(bytesleft)} left]"
+                raise SyntaxError(msg)
+            band = Image.frombuffer("L", pixel_size, b"".join(data), "raw", "L", 0, 1)
+            im.im.putband(band.im, band_ix)
+    return {"RGB": im}
+
+
+def read_mk(
+    fobj: IO[bytes], start_length: tuple[int, int], size: tuple[int, int, int]
+) -> dict[str, Image.Image]:
+    # Alpha masks seem to be uncompressed
+    start = start_length[0]
+    fobj.seek(start)
+    pixel_size = (size[0] * size[2], size[1] * size[2])
+    sizesq = pixel_size[0] * pixel_size[1]
+    band = Image.frombuffer("L", pixel_size, fobj.read(sizesq), "raw", "L", 0, 1)
+    return {"A": band}
+
+
+def read_png_or_jpeg2000(
+    fobj: IO[bytes], start_length: tuple[int, int], size: tuple[int, int, int]
+) -> dict[str, Image.Image]:
+    (start, length) = start_length
+    fobj.seek(start)
+    sig = fobj.read(12)
+
+    im: Image.Image
+    if sig[:8] == b"\x89PNG\x0d\x0a\x1a\x0a":
+        fobj.seek(start)
+        im = PngImagePlugin.PngImageFile(fobj)
+        Image._decompression_bomb_check(im.size)
+        return {"RGBA": im}
+    elif (
+        sig[:4] == b"\xff\x4f\xff\x51"
+        or sig[:4] == b"\x0d\x0a\x87\x0a"
+        or sig == b"\x00\x00\x00\x0cjP  \x0d\x0a\x87\x0a"
+    ):
+        if not enable_jpeg2k:
+            msg = (
+                "Unsupported icon subimage format (rebuild PIL "
+                "with JPEG 2000 support to fix this)"
+            )
+            raise ValueError(msg)
+        # j2k, jpc or j2c
+        fobj.seek(start)
+        jp2kstream = fobj.read(length)
+        f = io.BytesIO(jp2kstream)
+        im = Jpeg2KImagePlugin.Jpeg2KImageFile(f)
+        Image._decompression_bomb_check(im.size)
+        if im.mode != "RGBA":
+            im = im.convert("RGBA")
+        return {"RGBA": im}
+    else:
+        msg = "Unsupported icon subimage format"
+        raise ValueError(msg)
+
+
+class IcnsFile:
+    SIZES = {
+        (512, 512, 2): [(b"ic10", read_png_or_jpeg2000)],
+        (512, 512, 1): [(b"ic09", read_png_or_jpeg2000)],
+        (256, 256, 2): [(b"ic14", read_png_or_jpeg2000)],
+        (256, 256, 1): [(b"ic08", read_png_or_jpeg2000)],
+        (128, 128, 2): [(b"ic13", read_png_or_jpeg2000)],
+        (128, 128, 1): [
+            (b"ic07", read_png_or_jpeg2000),
+            (b"it32", read_32t),
+            (b"t8mk", read_mk),
+        ],
+        (64, 64, 1): [(b"icp6", read_png_or_jpeg2000)],
+        (32, 32, 2): [(b"ic12", read_png_or_jpeg2000)],
+        (48, 48, 1): [(b"ih32", read_32), (b"h8mk", read_mk)],
+        (32, 32, 1): [
+            (b"icp5", read_png_or_jpeg2000),
+            (b"il32", read_32),
+            (b"l8mk", read_mk),
+        ],
+        (16, 16, 2): [(b"ic11", read_png_or_jpeg2000)],
+        (16, 16, 1): [
+            (b"icp4", read_png_or_jpeg2000),
+            (b"is32", read_32),
+            (b"s8mk", read_mk),
+        ],
+    }
+
+    def __init__(self, fobj: IO[bytes]) -> None:
+        """
+        fobj is a file-like object as an icns resource
+        """
+        # signature : (start, length)
+        self.dct = {}
+        self.fobj = fobj
+        sig, filesize = nextheader(fobj)
+        if not _accept(sig):
+            msg = "not an icns file"
+            raise SyntaxError(msg)
+        i = HEADERSIZE
+        while i < filesize:
+            sig, blocksize = nextheader(fobj)
+            if blocksize <= 0:
+                msg = "invalid block header"
+                raise SyntaxError(msg)
+            i += HEADERSIZE
+            blocksize -= HEADERSIZE
+            self.dct[sig] = (i, blocksize)
+            fobj.seek(blocksize, io.SEEK_CUR)
+            i += blocksize
+
+    def itersizes(self) -> list[tuple[int, int, int]]:
+        sizes = []
+        for size, fmts in self.SIZES.items():
+            for fmt, reader in fmts:
+                if fmt in self.dct:
+                    sizes.append(size)
+                    break
+        return sizes
+
+    def bestsize(self) -> tuple[int, int, int]:
+        sizes = self.itersizes()
+        if not sizes:
+            msg = "No 32bit icon resources found"
+            raise SyntaxError(msg)
+        return max(sizes)
+
+    def dataforsize(self, size: tuple[int, int, int]) -> dict[str, Image.Image]:
+        """
+        Get an icon resource as {channel: array}.  Note that
+        the arrays are bottom-up like windows bitmaps and will likely
+        need to be flipped or transposed in some way.
+        """
+        dct = {}
+        for code, reader in self.SIZES[size]:
+            desc = self.dct.get(code)
+            if desc is not None:
+                dct.update(reader(self.fobj, desc, size))
+        return dct
+
+    def getimage(
+        self, size: tuple[int, int] | tuple[int, int, int] | None = None
+    ) -> Image.Image:
+        if size is None:
+            size = self.bestsize()
+        elif len(size) == 2:
+            size = (size[0], size[1], 1)
+        channels = self.dataforsize(size)
+
+        im = channels.get("RGBA")
+        if im:
+            return im
+
+        im = channels["RGB"].copy()
+        try:
+            im.putalpha(channels["A"])
+        except KeyError:
+            pass
+        return im
+
+
+##
+# Image plugin for Mac OS icons.
+
+
+class IcnsImageFile(ImageFile.ImageFile):
+    """
+    PIL image support for Mac OS .icns files.
+    Chooses the best resolution, but will possibly load
+    a different size image if you mutate the size attribute
+    before calling 'load'.
+
+    The info dictionary has a key 'sizes' that is a list
+    of sizes that the icns file has.
+    """
+
+    format = "ICNS"
+    format_description = "Mac OS icns resource"
+
+    def _open(self) -> None:
+        self.icns = IcnsFile(self.fp)
+        self._mode = "RGBA"
+        self.info["sizes"] = self.icns.itersizes()
+        self.best_size = self.icns.bestsize()
+        self.size = (
+            self.best_size[0] * self.best_size[2],
+            self.best_size[1] * self.best_size[2],
+        )
+
+    @property  # type: ignore[override]
+    def size(self) -> tuple[int, int] | tuple[int, int, int]:
+        return self._size
+
+    @size.setter
+    def size(self, value: tuple[int, int] | tuple[int, int, int]) -> None:
+        if len(value) == 3:
+            deprecate("Setting size to (width, height, scale)", 12, "load(scale)")
+            if value in self.info["sizes"]:
+                self._size = value  # type: ignore[assignment]
+                return
+        else:
+            # Check that a matching size exists,
+            # or that there is a scale that would create a size that matches
+            for size in self.info["sizes"]:
+                simple_size = size[0] * size[2], size[1] * size[2]
+                scale = simple_size[0] // value[0]
+                if simple_size[1] / value[1] == scale:
+                    self._size = value
+                    return
+        msg = "This is not one of the allowed sizes of this image"
+        raise ValueError(msg)
+
+    def load(self, scale: int | None = None) -> Image.core.PixelAccess | None:
+        if scale is not None or len(self.size) == 3:
+            if scale is None and len(self.size) == 3:
+                scale = self.size[2]
+            assert scale is not None
+            width, height = self.size[:2]
+            self.size = width * scale, height * scale
+            self.best_size = width, height, scale
+
+        px = Image.Image.load(self)
+        if self._im is not None and self.im.size == self.size:
+            # Already loaded
+            return px
+        self.load_prepare()
+        # This is likely NOT the best way to do it, but whatever.
+        im = self.icns.getimage(self.best_size)
+
+        # If this is a PNG or JPEG 2000, it won't be loaded yet
+        px = im.load()
+
+        self.im = im.im
+        self._mode = im.mode
+        self.size = im.size
+
+        return px
+
+
+def _save(im: Image.Image, fp: IO[bytes], filename: str | bytes) -> None:
+    """
+    Saves the image as a series of PNG files,
+    that are then combined into a .icns file.
+    """
+    if hasattr(fp, "flush"):
+        fp.flush()
+
+    sizes = {
+        b"ic07": 128,
+        b"ic08": 256,
+        b"ic09": 512,
+        b"ic10": 1024,
+        b"ic11": 32,
+        b"ic12": 64,
+        b"ic13": 256,
+        b"ic14": 512,
+    }
+    provided_images = {im.width: im for im in im.encoderinfo.get("append_images", [])}
+    size_streams = {}
+    for size in set(sizes.values()):
+        image = (
+            provided_images[size]
+            if size in provided_images
+            else im.resize((size, size))
+        )
+
+        temp = io.BytesIO()
+        image.save(temp, "png")
+        size_streams[size] = temp.getvalue()
+
+    entries = []
+    for type, size in sizes.items():
+        stream = size_streams[size]
+        entries.append((type, HEADERSIZE + len(stream), stream))
+
+    # Header
+    fp.write(MAGIC)
+    file_length = HEADERSIZE  # Header
+    file_length += HEADERSIZE + 8 * len(entries)  # TOC
+    file_length += sum(entry[1] for entry in entries)
+    fp.write(struct.pack(">i", file_length))
+
+    # TOC
+    fp.write(b"TOC ")
+    fp.write(struct.pack(">i", HEADERSIZE + len(entries) * HEADERSIZE))
+    for entry in entries:
+        fp.write(entry[0])
+        fp.write(struct.pack(">i", entry[1]))
+
+    # Data
+    for entry in entries:
+        fp.write(entry[0])
+        fp.write(struct.pack(">i", entry[1]))
+        fp.write(entry[2])
+
+    if hasattr(fp, "flush"):
+        fp.flush()
+
+
+def _accept(prefix: bytes) -> bool:
+    return prefix[:4] == MAGIC
+
+
+Image.register_open(IcnsImageFile.format, IcnsImageFile, _accept)
+Image.register_extension(IcnsImageFile.format, ".icns")
+
+Image.register_save(IcnsImageFile.format, _save)
+Image.register_mime(IcnsImageFile.format, "image/icns")
+
+if __name__ == "__main__":
+    if len(sys.argv) < 2:
+        print("Syntax: python3 IcnsImagePlugin.py [file]")
+        sys.exit()
+
+    with open(sys.argv[1], "rb") as fp:
+        imf = IcnsImageFile(fp)
+        for size in imf.info["sizes"]:
+            width, height, scale = imf.size = size
+            imf.save(f"out-{width}-{height}-{scale}.png")
+        with Image.open(sys.argv[1]) as im:
+            im.save("out.png")
+        if sys.platform == "windows":
+            os.startfile("out.png")

venv/Lib/site-packages/PIL/IcoImagePlugin.py

@@ -0,0 +1,381 @@
+#
+# The Python Imaging Library.
+# $Id$
+#
+# Windows Icon support for PIL
+#
+# History:
+#       96-05-27 fl     Created
+#
+# Copyright (c) Secret Labs AB 1997.
+# Copyright (c) Fredrik Lundh 1996.
+#
+# See the README file for information on usage and redistribution.
+#
+
+# This plugin is a refactored version of Win32IconImagePlugin by Bryan Davis
+# <casadebender@gmail.com>.
+# https://code.google.com/archive/p/casadebender/wikis/Win32IconImagePlugin.wiki
+#
+# Icon format references:
+#   * https://en.wikipedia.org/wiki/ICO_(file_format)
+#   * https://msdn.microsoft.com/en-us/library/ms997538.aspx
+from __future__ import annotations
+
+import warnings
+from io import BytesIO
+from math import ceil, log
+from typing import IO, NamedTuple
+
+from . import BmpImagePlugin, Image, ImageFile, PngImagePlugin
+from ._binary import i16le as i16
+from ._binary import i32le as i32
+from ._binary import o8
+from ._binary import o16le as o16
+from ._binary import o32le as o32
+
+#
+# --------------------------------------------------------------------
+
+_MAGIC = b"\0\0\1\0"
+
+
+def _save(im: Image.Image, fp: IO[bytes], filename: str | bytes) -> None:
+    fp.write(_MAGIC)  # (2+2)
+    bmp = im.encoderinfo.get("bitmap_format") == "bmp"
+    sizes = im.encoderinfo.get(
+        "sizes",
+        [(16, 16), (24, 24), (32, 32), (48, 48), (64, 64), (128, 128), (256, 256)],
+    )
+    frames = []
+    provided_ims = [im] + im.encoderinfo.get("append_images", [])
+    width, height = im.size
+    for size in sorted(set(sizes)):
+        if size[0] > width or size[1] > height or size[0] > 256 or size[1] > 256:
+            continue
+
+        for provided_im in provided_ims:
+            if provided_im.size != size:
+                continue
+            frames.append(provided_im)
+            if bmp:
+                bits = BmpImagePlugin.SAVE[provided_im.mode][1]
+                bits_used = [bits]
+                for other_im in provided_ims:
+                    if other_im.size != size:
+                        continue
+                    bits = BmpImagePlugin.SAVE[other_im.mode][1]
+                    if bits not in bits_used:
+                        # Another image has been supplied for this size
+                        # with a different bit depth
+                        frames.append(other_im)
+                        bits_used.append(bits)
+            break
+        else:
+            # TODO: invent a more convenient method for proportional scalings
+            frame = provided_im.copy()
+            frame.thumbnail(size, Image.Resampling.LANCZOS, reducing_gap=None)
+            frames.append(frame)
+    fp.write(o16(len(frames)))  # idCount(2)
+    offset = fp.tell() + len(frames) * 16
+    for frame in frames:
+        width, height = frame.size
+        # 0 means 256
+        fp.write(o8(width if width < 256 else 0))  # bWidth(1)
+        fp.write(o8(height if height < 256 else 0))  # bHeight(1)
+
+        bits, colors = BmpImagePlugin.SAVE[frame.mode][1:] if bmp else (32, 0)
+        fp.write(o8(colors))  # bColorCount(1)
+        fp.write(b"\0")  # bReserved(1)
+        fp.write(b"\0\0")  # wPlanes(2)
+        fp.write(o16(bits))  # wBitCount(2)
+
+        image_io = BytesIO()
+        if bmp:
+            frame.save(image_io, "dib")
+
+            if bits != 32:
+                and_mask = Image.new("1", size)
+                ImageFile._save(
+                    and_mask,
+                    image_io,
+                    [ImageFile._Tile("raw", (0, 0) + size, 0, ("1", 0, -1))],
+                )
+        else:
+            frame.save(image_io, "png")
+        image_io.seek(0)
+        image_bytes = image_io.read()
+        if bmp:
+            image_bytes = image_bytes[:8] + o32(height * 2) + image_bytes[12:]
+        bytes_len = len(image_bytes)
+        fp.write(o32(bytes_len))  # dwBytesInRes(4)
+        fp.write(o32(offset))  # dwImageOffset(4)
+        current = fp.tell()
+        fp.seek(offset)
+        fp.write(image_bytes)
+        offset = offset + bytes_len
+        fp.seek(current)
+
+
+def _accept(prefix: bytes) -> bool:
+    return prefix[:4] == _MAGIC
+
+
+class IconHeader(NamedTuple):
+    width: int
+    height: int
+    nb_color: int
+    reserved: int
+    planes: int
+    bpp: int
+    size: int
+    offset: int
+    dim: tuple[int, int]
+    square: int
+    color_depth: int
+
+
+class IcoFile:
+    def __init__(self, buf: IO[bytes]) -> None:
+        """
+        Parse image from file-like object containing ico file data
+        """
+
+        # check magic
+        s = buf.read(6)
+        if not _accept(s):
+            msg = "not an ICO file"
+            raise SyntaxError(msg)
+
+        self.buf = buf
+        self.entry = []
+
+        # Number of items in file
+        self.nb_items = i16(s, 4)
+
+        # Get headers for each item
+        for i in range(self.nb_items):
+            s = buf.read(16)
+
+            # See Wikipedia
+            width = s[0] or 256
+            height = s[1] or 256
+
+            # No. of colors in image (0 if >=8bpp)
+            nb_color = s[2]
+            bpp = i16(s, 6)
+            icon_header = IconHeader(
+                width=width,
+                height=height,
+                nb_color=nb_color,
+                reserved=s[3],
+                planes=i16(s, 4),
+                bpp=i16(s, 6),
+                size=i32(s, 8),
+                offset=i32(s, 12),
+                dim=(width, height),
+                square=width * height,
+                # See Wikipedia notes about color depth.
+                # We need this just to differ images with equal sizes
+                color_depth=bpp or (nb_color != 0 and ceil(log(nb_color, 2))) or 256,
+            )
+
+            self.entry.append(icon_header)
+
+        self.entry = sorted(self.entry, key=lambda x: x.color_depth)
+        # ICO images are usually squares
+        self.entry = sorted(self.entry, key=lambda x: x.square, reverse=True)
+
+    def sizes(self) -> set[tuple[int, int]]:
+        """
+        Get a set of all available icon sizes and color depths.
+        """
+        return {(h.width, h.height) for h in self.entry}
+
+    def getentryindex(self, size: tuple[int, int], bpp: int | bool = False) -> int:
+        for i, h in enumerate(self.entry):
+            if size == h.dim and (bpp is False or bpp == h.color_depth):
+                return i
+        return 0
+
+    def getimage(self, size: tuple[int, int], bpp: int | bool = False) -> Image.Image:
+        """
+        Get an image from the icon
+        """
+        return self.frame(self.getentryindex(size, bpp))
+
+    def frame(self, idx: int) -> Image.Image:
+        """
+        Get an image from frame idx
+        """
+
+        header = self.entry[idx]
+
+        self.buf.seek(header.offset)
+        data = self.buf.read(8)
+        self.buf.seek(header.offset)
+
+        im: Image.Image
+        if data[:8] == PngImagePlugin._MAGIC:
+            # png frame
+            im = PngImagePlugin.PngImageFile(self.buf)
+            Image._decompression_bomb_check(im.size)
+        else:
+            # XOR + AND mask bmp frame
+            im = BmpImagePlugin.DibImageFile(self.buf)
+            Image._decompression_bomb_check(im.size)
+
+            # change tile dimension to only encompass XOR image
+            im._size = (im.size[0], int(im.size[1] / 2))
+            d, e, o, a = im.tile[0]
+            im.tile[0] = ImageFile._Tile(d, (0, 0) + im.size, o, a)
+
+            # figure out where AND mask image starts
+            if header.bpp == 32:
+                # 32-bit color depth icon image allows semitransparent areas
+                # PIL's DIB format ignores transparency bits, recover them.
+                # The DIB is packed in BGRX byte order where X is the alpha
+                # channel.
+
+                # Back up to start of bmp data
+                self.buf.seek(o)
+                # extract every 4th byte (eg. 3,7,11,15,...)
+                alpha_bytes = self.buf.read(im.size[0] * im.size[1] * 4)[3::4]
+
+                # convert to an 8bpp grayscale image
+                try:
+                    mask = Image.frombuffer(
+                        "L",  # 8bpp
+                        im.size,  # (w, h)
+                        alpha_bytes,  # source chars
+                        "raw",  # raw decoder
+                        ("L", 0, -1),  # 8bpp inverted, unpadded, reversed
+                    )
+                except ValueError:
+                    if ImageFile.LOAD_TRUNCATED_IMAGES:
+                        mask = None
+                    else:
+                        raise
+            else:
+                # get AND image from end of bitmap
+                w = im.size[0]
+                if (w % 32) > 0:
+                    # bitmap row data is aligned to word boundaries
+                    w += 32 - (im.size[0] % 32)
+
+                # the total mask data is
+                # padded row size * height / bits per char
+
+                total_bytes = int((w * im.size[1]) / 8)
+                and_mask_offset = header.offset + header.size - total_bytes
+
+                self.buf.seek(and_mask_offset)
+                mask_data = self.buf.read(total_bytes)
+
+                # convert raw data to image
+                try:
+                    mask = Image.frombuffer(
+                        "1",  # 1 bpp
+                        im.size,  # (w, h)
+                        mask_data,  # source chars
+                        "raw",  # raw decoder
+                        ("1;I", int(w / 8), -1),  # 1bpp inverted, padded, reversed
+                    )
+                except ValueError:
+                    if ImageFile.LOAD_TRUNCATED_IMAGES:
+                        mask = None
+                    else:
+                        raise
+
+                # now we have two images, im is XOR image and mask is AND image
+
+            # apply mask image as alpha channel
+            if mask:
+                im = im.convert("RGBA")
+                im.putalpha(mask)
+
+        return im
+
+
+##
+# Image plugin for Windows Icon files.
+
+
+class IcoImageFile(ImageFile.ImageFile):
+    """
+    PIL read-only image support for Microsoft Windows .ico files.
+
+    By default the largest resolution image in the file will be loaded. This
+    can be changed by altering the 'size' attribute before calling 'load'.
+
+    The info dictionary has a key 'sizes' that is a list of the sizes available
+    in the icon file.
+
+    Handles classic, XP and Vista icon formats.
+
+    When saving, PNG compression is used. Support for this was only added in
+    Windows Vista. If you are unable to view the icon in Windows, convert the
+    image to "RGBA" mode before saving.
+
+    This plugin is a refactored version of Win32IconImagePlugin by Bryan Davis
+    <casadebender@gmail.com>.
+    https://code.google.com/archive/p/casadebender/wikis/Win32IconImagePlugin.wiki
+    """
+
+    format = "ICO"
+    format_description = "Windows Icon"
+
+    def _open(self) -> None:
+        self.ico = IcoFile(self.fp)
+        self.info["sizes"] = self.ico.sizes()
+        self.size = self.ico.entry[0].dim
+        self.load()
+
+    @property
+    def size(self) -> tuple[int, int]:
+        return self._size
+
+    @size.setter
+    def size(self, value: tuple[int, int]) -> None:
+        if value not in self.info["sizes"]:
+            msg = "This is not one of the allowed sizes of this image"
+            raise ValueError(msg)
+        self._size = value
+
+    def load(self) -> Image.core.PixelAccess | None:
+        if self._im is not None and self.im.size == self.size:
+            # Already loaded
+            return Image.Image.load(self)
+        im = self.ico.getimage(self.size)
+        # if tile is PNG, it won't really be loaded yet
+        im.load()
+        self.im = im.im
+        self._mode = im.mode
+        if im.palette:
+            self.palette = im.palette
+        if im.size != self.size:
+            warnings.warn("Image was not the expected size")
+
+            index = self.ico.getentryindex(self.size)
+            sizes = list(self.info["sizes"])
+            sizes[index] = im.size
+            self.info["sizes"] = set(sizes)
+
+            self.size = im.size
+        return None
+
+    def load_seek(self, pos: int) -> None:
+        # Flag the ImageFile.Parser so that it
+        # just does all the decode at the end.
+        pass
+
+
+#
+# --------------------------------------------------------------------
+
+
+Image.register_open(IcoImageFile.format, IcoImageFile, _accept)
+Image.register_save(IcoImageFile.format, _save)
+Image.register_extension(IcoImageFile.format, ".ico")
+
+Image.register_mime(IcoImageFile.format, "image/x-icon")

venv/Lib/site-packages/PIL/ImImagePlugin.py

@@ -0,0 +1,386 @@
+#
+# The Python Imaging Library.
+# $Id$
+#
+# IFUNC IM file handling for PIL
+#
+# history:
+# 1995-09-01 fl   Created.
+# 1997-01-03 fl   Save palette images
+# 1997-01-08 fl   Added sequence support
+# 1997-01-23 fl   Added P and RGB save support
+# 1997-05-31 fl   Read floating point images
+# 1997-06-22 fl   Save floating point images
+# 1997-08-27 fl   Read and save 1-bit images
+# 1998-06-25 fl   Added support for RGB+LUT images
+# 1998-07-02 fl   Added support for YCC images
+# 1998-07-15 fl   Renamed offset attribute to avoid name clash
+# 1998-12-29 fl   Added I;16 support
+# 2001-02-17 fl   Use 're' instead of 'regex' (Python 2.1) (0.7)
+# 2003-09-26 fl   Added LA/PA support
+#
+# Copyright (c) 1997-2003 by Secret Labs AB.
+# Copyright (c) 1995-2001 by Fredrik Lundh.
+#
+# See the README file for information on usage and redistribution.
+#
+from __future__ import annotations
+
+import os
+import re
+from typing import IO, Any
+
+from . import Image, ImageFile, ImagePalette
+
+# --------------------------------------------------------------------
+# Standard tags
+
+COMMENT = "Comment"
+DATE = "Date"
+EQUIPMENT = "Digitalization equipment"
+FRAMES = "File size (no of images)"
+LUT = "Lut"
+NAME = "Name"
+SCALE = "Scale (x,y)"
+SIZE = "Image size (x*y)"
+MODE = "Image type"
+
+TAGS = {
+    COMMENT: 0,
+    DATE: 0,
+    EQUIPMENT: 0,
+    FRAMES: 0,
+    LUT: 0,
+    NAME: 0,
+    SCALE: 0,
+    SIZE: 0,
+    MODE: 0,
+}
+
+OPEN = {
+    # ifunc93/p3cfunc formats
+    "0 1 image": ("1", "1"),
+    "L 1 image": ("1", "1"),
+    "Greyscale image": ("L", "L"),
+    "Grayscale image": ("L", "L"),
+    "RGB image": ("RGB", "RGB;L"),
+    "RLB image": ("RGB", "RLB"),
+    "RYB image": ("RGB", "RLB"),
+    "B1 image": ("1", "1"),
+    "B2 image": ("P", "P;2"),
+    "B4 image": ("P", "P;4"),
+    "X 24 image": ("RGB", "RGB"),
+    "L 32 S image": ("I", "I;32"),
+    "L 32 F image": ("F", "F;32"),
+    # old p3cfunc formats
+    "RGB3 image": ("RGB", "RGB;T"),
+    "RYB3 image": ("RGB", "RYB;T"),
+    # extensions
+    "LA image": ("LA", "LA;L"),
+    "PA image": ("LA", "PA;L"),
+    "RGBA image": ("RGBA", "RGBA;L"),
+    "RGBX image": ("RGB", "RGBX;L"),
+    "CMYK image": ("CMYK", "CMYK;L"),
+    "YCC image": ("YCbCr", "YCbCr;L"),
+}
+
+# ifunc95 extensions
+for i in ["8", "8S", "16", "16S", "32", "32F"]:
+    OPEN[f"L {i} image"] = ("F", f"F;{i}")
+    OPEN[f"L*{i} image"] = ("F", f"F;{i}")
+for i in ["16", "16L", "16B"]:
+    OPEN[f"L {i} image"] = (f"I;{i}", f"I;{i}")
+    OPEN[f"L*{i} image"] = (f"I;{i}", f"I;{i}")
+for i in ["32S"]:
+    OPEN[f"L {i} image"] = ("I", f"I;{i}")
+    OPEN[f"L*{i} image"] = ("I", f"I;{i}")
+for j in range(2, 33):
+    OPEN[f"L*{j} image"] = ("F", f"F;{j}")
+
+
+# --------------------------------------------------------------------
+# Read IM directory
+
+split = re.compile(rb"^([A-Za-z][^:]*):[ \t]*(.*)[ \t]*$")
+
+
+def number(s: Any) -> float:
+    try:
+        return int(s)
+    except ValueError:
+        return float(s)
+
+
+##
+# Image plugin for the IFUNC IM file format.
+
+
+class ImImageFile(ImageFile.ImageFile):
+    format = "IM"
+    format_description = "IFUNC Image Memory"
+    _close_exclusive_fp_after_loading = False
+
+    def _open(self) -> None:
+        # Quick rejection: if there's not an LF among the first
+        # 100 bytes, this is (probably) not a text header.
+
+        if b"\n" not in self.fp.read(100):
+            msg = "not an IM file"
+            raise SyntaxError(msg)
+        self.fp.seek(0)
+
+        n = 0
+
+        # Default values
+        self.info[MODE] = "L"
+        self.info[SIZE] = (512, 512)
+        self.info[FRAMES] = 1
+
+        self.rawmode = "L"
+
+        while True:
+            s = self.fp.read(1)
+
+            # Some versions of IFUNC uses \n\r instead of \r\n...
+            if s == b"\r":
+                continue
+
+            if not s or s == b"\0" or s == b"\x1A":
+                break
+
+            # FIXME: this may read whole file if not a text file
+            s = s + self.fp.readline()
+
+            if len(s) > 100:
+                msg = "not an IM file"
+                raise SyntaxError(msg)
+
+            if s[-2:] == b"\r\n":
+                s = s[:-2]
+            elif s[-1:] == b"\n":
+                s = s[:-1]
+
+            try:
+                m = split.match(s)
+            except re.error as e:
+                msg = "not an IM file"
+                raise SyntaxError(msg) from e
+
+            if m:
+                k, v = m.group(1, 2)
+
+                # Don't know if this is the correct encoding,
+                # but a decent guess (I guess)
+                k = k.decode("latin-1", "replace")
+                v = v.decode("latin-1", "replace")
+
+                # Convert value as appropriate
+                if k in [FRAMES, SCALE, SIZE]:
+                    v = v.replace("*", ",")
+                    v = tuple(map(number, v.split(",")))
+                    if len(v) == 1:
+                        v = v[0]
+                elif k == MODE and v in OPEN:
+                    v, self.rawmode = OPEN[v]
+
+                # Add to dictionary. Note that COMMENT tags are
+                # combined into a list of strings.
+                if k == COMMENT:
+                    if k in self.info:
+                        self.info[k].append(v)
+                    else:
+                        self.info[k] = [v]
+                else:
+                    self.info[k] = v
+
+                if k in TAGS:
+                    n += 1
+
+            else:
+                msg = f"Syntax error in IM header: {s.decode('ascii', 'replace')}"
+                raise SyntaxError(msg)
+
+        if not n:
+            msg = "Not an IM file"
+            raise SyntaxError(msg)
+
+        # Basic attributes
+        self._size = self.info[SIZE]
+        self._mode = self.info[MODE]
+
+        # Skip forward to start of image data
+        while s and s[:1] != b"\x1A":
+            s = self.fp.read(1)
+        if not s:
+            msg = "File truncated"
+            raise SyntaxError(msg)
+
+        if LUT in self.info:
+            # convert lookup table to palette or lut attribute
+            palette = self.fp.read(768)
+            greyscale = 1  # greyscale palette
+            linear = 1  # linear greyscale palette
+            for i in range(256):
+                if palette[i] == palette[i + 256] == palette[i + 512]:
+                    if palette[i] != i:
+                        linear = 0
+                else:
+                    greyscale = 0
+            if self.mode in ["L", "LA", "P", "PA"]:
+                if greyscale:
+                    if not linear:
+                        self.lut = list(palette[:256])
+                else:
+                    if self.mode in ["L", "P"]:
+                        self._mode = self.rawmode = "P"
+                    elif self.mode in ["LA", "PA"]:
+                        self._mode = "PA"
+                        self.rawmode = "PA;L"
+                    self.palette = ImagePalette.raw("RGB;L", palette)
+            elif self.mode == "RGB":
+                if not greyscale or not linear:
+                    self.lut = list(palette)
+
+        self.frame = 0
+
+        self.__offset = offs = self.fp.tell()
+
+        self._fp = self.fp  # FIXME: hack
+
+        if self.rawmode[:2] == "F;":
+            # ifunc95 formats
+            try:
+                # use bit decoder (if necessary)
+                bits = int(self.rawmode[2:])
+                if bits not in [8, 16, 32]:
+                    self.tile = [
+                        ImageFile._Tile(
+                            "bit", (0, 0) + self.size, offs, (bits, 8, 3, 0, -1)
+                        )
+                    ]
+                    return
+            except ValueError:
+                pass
+
+        if self.rawmode in ["RGB;T", "RYB;T"]:
+            # Old LabEye/3PC files.  Would be very surprised if anyone
+            # ever stumbled upon such a file ;-)
+            size = self.size[0] * self.size[1]
+            self.tile = [
+                ImageFile._Tile("raw", (0, 0) + self.size, offs, ("G", 0, -1)),
+                ImageFile._Tile("raw", (0, 0) + self.size, offs + size, ("R", 0, -1)),
+                ImageFile._Tile(
+                    "raw", (0, 0) + self.size, offs + 2 * size, ("B", 0, -1)
+                ),
+            ]
+        else:
+            # LabEye/IFUNC files
+            self.tile = [
+                ImageFile._Tile("raw", (0, 0) + self.size, offs, (self.rawmode, 0, -1))
+            ]
+
+    @property
+    def n_frames(self) -> int:
+        return self.info[FRAMES]
+
+    @property
+    def is_animated(self) -> bool:
+        return self.info[FRAMES] > 1
+
+    def seek(self, frame: int) -> None:
+        if not self._seek_check(frame):
+            return
+
+        self.frame = frame
+
+        if self.mode == "1":
+            bits = 1
+        else:
+            bits = 8 * len(self.mode)
+
+        size = ((self.size[0] * bits + 7) // 8) * self.size[1]
+        offs = self.__offset + frame * size
+
+        self.fp = self._fp
+
+        self.tile = [
+            ImageFile._Tile("raw", (0, 0) + self.size, offs, (self.rawmode, 0, -1))
+        ]
+
+    def tell(self) -> int:
+        return self.frame
+
+
+#
+# --------------------------------------------------------------------
+# Save IM files
+
+
+SAVE = {
+    # mode: (im type, raw mode)
+    "1": ("0 1", "1"),
+    "L": ("Greyscale", "L"),
+    "LA": ("LA", "LA;L"),
+    "P": ("Greyscale", "P"),
+    "PA": ("LA", "PA;L"),
+    "I": ("L 32S", "I;32S"),
+    "I;16": ("L 16", "I;16"),
+    "I;16L": ("L 16L", "I;16L"),
+    "I;16B": ("L 16B", "I;16B"),
+    "F": ("L 32F", "F;32F"),
+    "RGB": ("RGB", "RGB;L"),
+    "RGBA": ("RGBA", "RGBA;L"),
+    "RGBX": ("RGBX", "RGBX;L"),
+    "CMYK": ("CMYK", "CMYK;L"),
+    "YCbCr": ("YCC", "YCbCr;L"),
+}
+
+
+def _save(im: Image.Image, fp: IO[bytes], filename: str | bytes) -> None:
+    try:
+        image_type, rawmode = SAVE[im.mode]
+    except KeyError as e:
+        msg = f"Cannot save {im.mode} images as IM"
+        raise ValueError(msg) from e
+
+    frames = im.encoderinfo.get("frames", 1)
+
+    fp.write(f"Image type: {image_type} image\r\n".encode("ascii"))
+    if filename:
+        # Each line must be 100 characters or less,
+        # or: SyntaxError("not an IM file")
+        # 8 characters are used for "Name: " and "\r\n"
+        # Keep just the filename, ditch the potentially overlong path
+        if isinstance(filename, bytes):
+            filename = filename.decode("ascii")
+        name, ext = os.path.splitext(os.path.basename(filename))
+        name = "".join([name[: 92 - len(ext)], ext])
+
+        fp.write(f"Name: {name}\r\n".encode("ascii"))
+    fp.write(f"Image size (x*y): {im.size[0]}*{im.size[1]}\r\n".encode("ascii"))
+    fp.write(f"File size (no of images): {frames}\r\n".encode("ascii"))
+    if im.mode in ["P", "PA"]:
+        fp.write(b"Lut: 1\r\n")
+    fp.write(b"\000" * (511 - fp.tell()) + b"\032")
+    if im.mode in ["P", "PA"]:
+        im_palette = im.im.getpalette("RGB", "RGB;L")
+        colors = len(im_palette) // 3
+        palette = b""
+        for i in range(3):
+            palette += im_palette[colors * i : colors * (i + 1)]
+            palette += b"\x00" * (256 - colors)
+        fp.write(palette)  # 768 bytes
+    ImageFile._save(
+        im, fp, [ImageFile._Tile("raw", (0, 0) + im.size, 0, (rawmode, 0, -1))]
+    )
+
+
+#
+# --------------------------------------------------------------------
+# Registry
+
+
+Image.register_open(ImImageFile.format, ImImageFile)
+Image.register_save(ImImageFile.format, _save)
+
+Image.register_extension(ImImageFile.format, ".im")

venv/Lib/site-packages/PIL/ImageChops.py

@@ -0,0 +1,311 @@
+#
+# The Python Imaging Library.
+# $Id$
+#
+# standard channel operations
+#
+# History:
+# 1996-03-24 fl   Created
+# 1996-08-13 fl   Added logical operations (for "1" images)
+# 2000-10-12 fl   Added offset method (from Image.py)
+#
+# Copyright (c) 1997-2000 by Secret Labs AB
+# Copyright (c) 1996-2000 by Fredrik Lundh
+#
+# See the README file for information on usage and redistribution.
+#
+
+from __future__ import annotations
+
+from . import Image
+
+
+def constant(image: Image.Image, value: int) -> Image.Image:
+    """Fill a channel with a given gray level.
+
+    :rtype: :py:class:`~PIL.Image.Image`
+    """
+
+    return Image.new("L", image.size, value)
+
+
+def duplicate(image: Image.Image) -> Image.Image:
+    """Copy a channel. Alias for :py:meth:`PIL.Image.Image.copy`.
+
+    :rtype: :py:class:`~PIL.Image.Image`
+    """
+
+    return image.copy()
+
+
+def invert(image: Image.Image) -> Image.Image:
+    """
+    Invert an image (channel). ::
+
+        out = MAX - image
+
+    :rtype: :py:class:`~PIL.Image.Image`
+    """
+
+    image.load()
+    return image._new(image.im.chop_invert())
+
+
+def lighter(image1: Image.Image, image2: Image.Image) -> Image.Image:
+    """
+    Compares the two images, pixel by pixel, and returns a new image containing
+    the lighter values. ::
+
+        out = max(image1, image2)
+
+    :rtype: :py:class:`~PIL.Image.Image`
+    """
+
+    image1.load()
+    image2.load()
+    return image1._new(image1.im.chop_lighter(image2.im))
+
+
+def darker(image1: Image.Image, image2: Image.Image) -> Image.Image:
+    """
+    Compares the two images, pixel by pixel, and returns a new image containing
+    the darker values. ::
+
+        out = min(image1, image2)
+
+    :rtype: :py:class:`~PIL.Image.Image`
+    """
+
+    image1.load()
+    image2.load()
+    return image1._new(image1.im.chop_darker(image2.im))
+
+
+def difference(image1: Image.Image, image2: Image.Image) -> Image.Image:
+    """
+    Returns the absolute value of the pixel-by-pixel difference between the two
+    images. ::
+
+        out = abs(image1 - image2)
+
+    :rtype: :py:class:`~PIL.Image.Image`
+    """
+
+    image1.load()
+    image2.load()
+    return image1._new(image1.im.chop_difference(image2.im))
+
+
+def multiply(image1: Image.Image, image2: Image.Image) -> Image.Image:
+    """
+    Superimposes two images on top of each other.
+
+    If you multiply an image with a solid black image, the result is black. If
+    you multiply with a solid white image, the image is unaffected. ::
+
+        out = image1 * image2 / MAX
+
+    :rtype: :py:class:`~PIL.Image.Image`
+    """
+
+    image1.load()
+    image2.load()
+    return image1._new(image1.im.chop_multiply(image2.im))
+
+
+def screen(image1: Image.Image, image2: Image.Image) -> Image.Image:
+    """
+    Superimposes two inverted images on top of each other. ::
+
+        out = MAX - ((MAX - image1) * (MAX - image2) / MAX)
+
+    :rtype: :py:class:`~PIL.Image.Image`
+    """
+
+    image1.load()
+    image2.load()
+    return image1._new(image1.im.chop_screen(image2.im))
+
+
+def soft_light(image1: Image.Image, image2: Image.Image) -> Image.Image:
+    """
+    Superimposes two images on top of each other using the Soft Light algorithm
+
+    :rtype: :py:class:`~PIL.Image.Image`
+    """
+
+    image1.load()
+    image2.load()
+    return image1._new(image1.im.chop_soft_light(image2.im))
+
+
+def hard_light(image1: Image.Image, image2: Image.Image) -> Image.Image:
+    """
+    Superimposes two images on top of each other using the Hard Light algorithm
+
+    :rtype: :py:class:`~PIL.Image.Image`
+    """
+
+    image1.load()
+    image2.load()
+    return image1._new(image1.im.chop_hard_light(image2.im))
+
+
+def overlay(image1: Image.Image, image2: Image.Image) -> Image.Image:
+    """
+    Superimposes two images on top of each other using the Overlay algorithm
+
+    :rtype: :py:class:`~PIL.Image.Image`
+    """
+
+    image1.load()
+    image2.load()
+    return image1._new(image1.im.chop_overlay(image2.im))
+
+
+def add(
+    image1: Image.Image, image2: Image.Image, scale: float = 1.0, offset: float = 0
+) -> Image.Image:
+    """
+    Adds two images, dividing the result by scale and adding the
+    offset. If omitted, scale defaults to 1.0, and offset to 0.0. ::
+
+        out = ((image1 + image2) / scale + offset)
+
+    :rtype: :py:class:`~PIL.Image.Image`
+    """
+
+    image1.load()
+    image2.load()
+    return image1._new(image1.im.chop_add(image2.im, scale, offset))
+
+
+def subtract(
+    image1: Image.Image, image2: Image.Image, scale: float = 1.0, offset: float = 0
+) -> Image.Image:
+    """
+    Subtracts two images, dividing the result by scale and adding the offset.
+    If omitted, scale defaults to 1.0, and offset to 0.0. ::
+
+        out = ((image1 - image2) / scale + offset)
+
+    :rtype: :py:class:`~PIL.Image.Image`
+    """
+
+    image1.load()
+    image2.load()
+    return image1._new(image1.im.chop_subtract(image2.im, scale, offset))
+
+
+def add_modulo(image1: Image.Image, image2: Image.Image) -> Image.Image:
+    """Add two images, without clipping the result. ::
+
+        out = ((image1 + image2) % MAX)
+
+    :rtype: :py:class:`~PIL.Image.Image`
+    """
+
+    image1.load()
+    image2.load()
+    return image1._new(image1.im.chop_add_modulo(image2.im))
+
+
+def subtract_modulo(image1: Image.Image, image2: Image.Image) -> Image.Image:
+    """Subtract two images, without clipping the result. ::
+
+        out = ((image1 - image2) % MAX)
+
+    :rtype: :py:class:`~PIL.Image.Image`
+    """
+
+    image1.load()
+    image2.load()
+    return image1._new(image1.im.chop_subtract_modulo(image2.im))
+
+
+def logical_and(image1: Image.Image, image2: Image.Image) -> Image.Image:
+    """Logical AND between two images.
+
+    Both of the images must have mode "1". If you would like to perform a
+    logical AND on an image with a mode other than "1", try
+    :py:meth:`~PIL.ImageChops.multiply` instead, using a black-and-white mask
+    as the second image. ::
+
+        out = ((image1 and image2) % MAX)
+
+    :rtype: :py:class:`~PIL.Image.Image`
+    """
+
+    image1.load()
+    image2.load()
+    return image1._new(image1.im.chop_and(image2.im))
+
+
+def logical_or(image1: Image.Image, image2: Image.Image) -> Image.Image:
+    """Logical OR between two images.
+
+    Both of the images must have mode "1". ::
+
+        out = ((image1 or image2) % MAX)
+
+    :rtype: :py:class:`~PIL.Image.Image`
+    """
+
+    image1.load()
+    image2.load()
+    return image1._new(image1.im.chop_or(image2.im))
+
+
+def logical_xor(image1: Image.Image, image2: Image.Image) -> Image.Image:
+    """Logical XOR between two images.
+
+    Both of the images must have mode "1". ::
+
+        out = ((bool(image1) != bool(image2)) % MAX)
+
+    :rtype: :py:class:`~PIL.Image.Image`
+    """
+
+    image1.load()
+    image2.load()
+    return image1._new(image1.im.chop_xor(image2.im))
+
+
+def blend(image1: Image.Image, image2: Image.Image, alpha: float) -> Image.Image:
+    """Blend images using constant transparency weight. Alias for
+    :py:func:`PIL.Image.blend`.
+
+    :rtype: :py:class:`~PIL.Image.Image`
+    """
+
+    return Image.blend(image1, image2, alpha)
+
+
+def composite(
+    image1: Image.Image, image2: Image.Image, mask: Image.Image
+) -> Image.Image:
+    """Create composite using transparency mask. Alias for
+    :py:func:`PIL.Image.composite`.
+
+    :rtype: :py:class:`~PIL.Image.Image`
+    """
+
+    return Image.composite(image1, image2, mask)
+
+
+def offset(image: Image.Image, xoffset: int, yoffset: int | None = None) -> Image.Image:
+    """Returns a copy of the image where data has been offset by the given
+    distances. Data wraps around the edges. If ``yoffset`` is omitted, it
+    is assumed to be equal to ``xoffset``.
+
+    :param image: Input image.
+    :param xoffset: The horizontal distance.
+    :param yoffset: The vertical distance.  If omitted, both
+        distances are set to the same value.
+    :rtype: :py:class:`~PIL.Image.Image`
+    """
+
+    if yoffset is None:
+        yoffset = xoffset
+    image.load()
+    return image._new(image.im.offset(xoffset, yoffset))

venv/Lib/site-packages/PIL/ImageColor.py

@@ -0,0 +1,320 @@
+#
+# The Python Imaging Library
+# $Id$
+#
+# map CSS3-style colour description strings to RGB
+#
+# History:
+# 2002-10-24 fl   Added support for CSS-style color strings
+# 2002-12-15 fl   Added RGBA support
+# 2004-03-27 fl   Fixed remaining int() problems for Python 1.5.2
+# 2004-07-19 fl   Fixed gray/grey spelling issues
+# 2009-03-05 fl   Fixed rounding error in grayscale calculation
+#
+# Copyright (c) 2002-2004 by Secret Labs AB
+# Copyright (c) 2002-2004 by Fredrik Lundh
+#
+# See the README file for information on usage and redistribution.
+#
+from __future__ import annotations
+
+import re
+from functools import lru_cache
+
+from . import Image
+
+
+@lru_cache
+def getrgb(color: str) -> tuple[int, int, int] | tuple[int, int, int, int]:
+    """
+     Convert a color string to an RGB or RGBA tuple. If the string cannot be
+     parsed, this function raises a :py:exc:`ValueError` exception.
+
+    .. versionadded:: 1.1.4
+
+    :param color: A color string
+    :return: ``(red, green, blue[, alpha])``
+    """
+    if len(color) > 100:
+        msg = "color specifier is too long"
+        raise ValueError(msg)
+    color = color.lower()
+
+    rgb = colormap.get(color, None)
+    if rgb:
+        if isinstance(rgb, tuple):
+            return rgb
+        rgb_tuple = getrgb(rgb)
+        assert len(rgb_tuple) == 3
+        colormap[color] = rgb_tuple
+        return rgb_tuple
+
+    # check for known string formats
+    if re.match("#[a-f0-9]{3}$", color):
+        return int(color[1] * 2, 16), int(color[2] * 2, 16), int(color[3] * 2, 16)
+
+    if re.match("#[a-f0-9]{4}$", color):
+        return (
+            int(color[1] * 2, 16),
+            int(color[2] * 2, 16),
+            int(color[3] * 2, 16),
+            int(color[4] * 2, 16),
+        )
+
+    if re.match("#[a-f0-9]{6}$", color):
+        return int(color[1:3], 16), int(color[3:5], 16), int(color[5:7], 16)
+
+    if re.match("#[a-f0-9]{8}$", color):
+        return (
+            int(color[1:3], 16),
+            int(color[3:5], 16),
+            int(color[5:7], 16),
+            int(color[7:9], 16),
+        )
+
+    m = re.match(r"rgb\(\s*(\d+)\s*,\s*(\d+)\s*,\s*(\d+)\s*\)$", color)
+    if m:
+        return int(m.group(1)), int(m.group(2)), int(m.group(3))
+
+    m = re.match(r"rgb\(\s*(\d+)%\s*,\s*(\d+)%\s*,\s*(\d+)%\s*\)$", color)
+    if m:
+        return (
+            int((int(m.group(1)) * 255) / 100.0 + 0.5),
+            int((int(m.group(2)) * 255) / 100.0 + 0.5),
+            int((int(m.group(3)) * 255) / 100.0 + 0.5),
+        )
+
+    m = re.match(
+        r"hsl\(\s*(\d+\.?\d*)\s*,\s*(\d+\.?\d*)%\s*,\s*(\d+\.?\d*)%\s*\)$", color
+    )
+    if m:
+        from colorsys import hls_to_rgb
+
+        rgb_floats = hls_to_rgb(
+            float(m.group(1)) / 360.0,
+            float(m.group(3)) / 100.0,
+            float(m.group(2)) / 100.0,
+        )
+        return (
+            int(rgb_floats[0] * 255 + 0.5),
+            int(rgb_floats[1] * 255 + 0.5),
+            int(rgb_floats[2] * 255 + 0.5),
+        )
+
+    m = re.match(
+        r"hs[bv]\(\s*(\d+\.?\d*)\s*,\s*(\d+\.?\d*)%\s*,\s*(\d+\.?\d*)%\s*\)$", color
+    )
+    if m:
+        from colorsys import hsv_to_rgb
+
+        rgb_floats = hsv_to_rgb(
+            float(m.group(1)) / 360.0,
+            float(m.group(2)) / 100.0,
+            float(m.group(3)) / 100.0,
+        )
+        return (
+            int(rgb_floats[0] * 255 + 0.5),
+            int(rgb_floats[1] * 255 + 0.5),
+            int(rgb_floats[2] * 255 + 0.5),
+        )
+
+    m = re.match(r"rgba\(\s*(\d+)\s*,\s*(\d+)\s*,\s*(\d+)\s*,\s*(\d+)\s*\)$", color)
+    if m:
+        return int(m.group(1)), int(m.group(2)), int(m.group(3)), int(m.group(4))
+    msg = f"unknown color specifier: {repr(color)}"
+    raise ValueError(msg)
+
+
+@lru_cache
+def getcolor(color: str, mode: str) -> int | tuple[int, ...]:
+    """
+    Same as :py:func:`~PIL.ImageColor.getrgb` for most modes. However, if
+    ``mode`` is HSV, converts the RGB value to a HSV value, or if ``mode`` is
+    not color or a palette image, converts the RGB value to a grayscale value.
+    If the string cannot be parsed, this function raises a :py:exc:`ValueError`
+    exception.
+
+    .. versionadded:: 1.1.4
+
+    :param color: A color string
+    :param mode: Convert result to this mode
+    :return: ``graylevel, (graylevel, alpha) or (red, green, blue[, alpha])``
+    """
+    # same as getrgb, but converts the result to the given mode
+    rgb, alpha = getrgb(color), 255
+    if len(rgb) == 4:
+        alpha = rgb[3]
+        rgb = rgb[:3]
+
+    if mode == "HSV":
+        from colorsys import rgb_to_hsv
+
+        r, g, b = rgb
+        h, s, v = rgb_to_hsv(r / 255, g / 255, b / 255)
+        return int(h * 255), int(s * 255), int(v * 255)
+    elif Image.getmodebase(mode) == "L":
+        r, g, b = rgb
+        # ITU-R Recommendation 601-2 for nonlinear RGB
+        # scaled to 24 bits to match the convert's implementation.
+        graylevel = (r * 19595 + g * 38470 + b * 7471 + 0x8000) >> 16
+        if mode[-1] == "A":
+            return graylevel, alpha
+        return graylevel
+    elif mode[-1] == "A":
+        return rgb + (alpha,)
+    return rgb
+
+
+colormap: dict[str, str | tuple[int, int, int]] = {
+    # X11 colour table from https://drafts.csswg.org/css-color-4/, with
+    # gray/grey spelling issues fixed.  This is a superset of HTML 4.0
+    # colour names used in CSS 1.
+    "aliceblue": "#f0f8ff",
+    "antiquewhite": "#faebd7",
+    "aqua": "#00ffff",
+    "aquamarine": "#7fffd4",
+    "azure": "#f0ffff",
+    "beige": "#f5f5dc",
+    "bisque": "#ffe4c4",
+    "black": "#000000",
+    "blanchedalmond": "#ffebcd",
+    "blue": "#0000ff",
+    "blueviolet": "#8a2be2",
+    "brown": "#a52a2a",
+    "burlywood": "#deb887",
+    "cadetblue": "#5f9ea0",
+    "chartreuse": "#7fff00",
+    "chocolate": "#d2691e",
+    "coral": "#ff7f50",
+    "cornflowerblue": "#6495ed",
+    "cornsilk": "#fff8dc",
+    "crimson": "#dc143c",
+    "cyan": "#00ffff",
+    "darkblue": "#00008b",
+    "darkcyan": "#008b8b",
+    "darkgoldenrod": "#b8860b",
+    "darkgray": "#a9a9a9",
+    "darkgrey": "#a9a9a9",
+    "darkgreen": "#006400",
+    "darkkhaki": "#bdb76b",
+    "darkmagenta": "#8b008b",
+    "darkolivegreen": "#556b2f",
+    "darkorange": "#ff8c00",
+    "darkorchid": "#9932cc",
+    "darkred": "#8b0000",
+    "darksalmon": "#e9967a",
+    "darkseagreen": "#8fbc8f",
+    "darkslateblue": "#483d8b",
+    "darkslategray": "#2f4f4f",
+    "darkslategrey": "#2f4f4f",
+    "darkturquoise": "#00ced1",
+    "darkviolet": "#9400d3",
+    "deeppink": "#ff1493",
+    "deepskyblue": "#00bfff",
+    "dimgray": "#696969",
+    "dimgrey": "#696969",
+    "dodgerblue": "#1e90ff",
+    "firebrick": "#b22222",
+    "floralwhite": "#fffaf0",
+    "forestgreen": "#228b22",
+    "fuchsia": "#ff00ff",
+    "gainsboro": "#dcdcdc",
+    "ghostwhite": "#f8f8ff",
+    "gold": "#ffd700",
+    "goldenrod": "#daa520",
+    "gray": "#808080",
+    "grey": "#808080",
+    "green": "#008000",
+    "greenyellow": "#adff2f",
+    "honeydew": "#f0fff0",
+    "hotpink": "#ff69b4",
+    "indianred": "#cd5c5c",
+    "indigo": "#4b0082",
+    "ivory": "#fffff0",
+    "khaki": "#f0e68c",
+    "lavender": "#e6e6fa",
+    "lavenderblush": "#fff0f5",
+    "lawngreen": "#7cfc00",
+    "lemonchiffon": "#fffacd",
+    "lightblue": "#add8e6",
+    "lightcoral": "#f08080",
+    "lightcyan": "#e0ffff",
+    "lightgoldenrodyellow": "#fafad2",
+    "lightgreen": "#90ee90",
+    "lightgray": "#d3d3d3",
+    "lightgrey": "#d3d3d3",
+    "lightpink": "#ffb6c1",
+    "lightsalmon": "#ffa07a",
+    "lightseagreen": "#20b2aa",
+    "lightskyblue": "#87cefa",
+    "lightslategray": "#778899",
+    "lightslategrey": "#778899",
+    "lightsteelblue": "#b0c4de",
+    "lightyellow": "#ffffe0",
+    "lime": "#00ff00",
+    "limegreen": "#32cd32",
+    "linen": "#faf0e6",
+    "magenta": "#ff00ff",
+    "maroon": "#800000",
+    "mediumaquamarine": "#66cdaa",
+    "mediumblue": "#0000cd",
+    "mediumorchid": "#ba55d3",
+    "mediumpurple": "#9370db",
+    "mediumseagreen": "#3cb371",
+    "mediumslateblue": "#7b68ee",
+    "mediumspringgreen": "#00fa9a",
+    "mediumturquoise": "#48d1cc",
+    "mediumvioletred": "#c71585",
+    "midnightblue": "#191970",
+    "mintcream": "#f5fffa",
+    "mistyrose": "#ffe4e1",
+    "moccasin": "#ffe4b5",
+    "navajowhite": "#ffdead",
+    "navy": "#000080",
+    "oldlace": "#fdf5e6",
+    "olive": "#808000",
+    "olivedrab": "#6b8e23",
+    "orange": "#ffa500",
+    "orangered": "#ff4500",
+    "orchid": "#da70d6",
+    "palegoldenrod": "#eee8aa",
+    "palegreen": "#98fb98",
+    "paleturquoise": "#afeeee",
+    "palevioletred": "#db7093",
+    "papayawhip": "#ffefd5",
+    "peachpuff": "#ffdab9",
+    "peru": "#cd853f",
+    "pink": "#ffc0cb",
+    "plum": "#dda0dd",
+    "powderblue": "#b0e0e6",
+    "purple": "#800080",
+    "rebeccapurple": "#663399",
+    "red": "#ff0000",
+    "rosybrown": "#bc8f8f",
+    "royalblue": "#4169e1",
+    "saddlebrown": "#8b4513",
+    "salmon": "#fa8072",
+    "sandybrown": "#f4a460",
+    "seagreen": "#2e8b57",
+    "seashell": "#fff5ee",
+    "sienna": "#a0522d",
+    "silver": "#c0c0c0",
+    "skyblue": "#87ceeb",
+    "slateblue": "#6a5acd",
+    "slategray": "#708090",
+    "slategrey": "#708090",
+    "snow": "#fffafa",
+    "springgreen": "#00ff7f",
+    "steelblue": "#4682b4",
+    "tan": "#d2b48c",
+    "teal": "#008080",
+    "thistle": "#d8bfd8",
+    "tomato": "#ff6347",
+    "turquoise": "#40e0d0",
+    "violet": "#ee82ee",
+    "wheat": "#f5deb3",
+    "white": "#ffffff",
+    "whitesmoke": "#f5f5f5",
+    "yellow": "#ffff00",
+    "yellowgreen": "#9acd32",
+}

venv/Lib/site-packages/PIL/ImageDraw2.py

@@ -0,0 +1,243 @@
+#
+# The Python Imaging Library
+# $Id$
+#
+# WCK-style drawing interface operations
+#
+# History:
+# 2003-12-07 fl   created
+# 2005-05-15 fl   updated; added to PIL as ImageDraw2
+# 2005-05-15 fl   added text support
+# 2005-05-20 fl   added arc/chord/pieslice support
+#
+# Copyright (c) 2003-2005 by Secret Labs AB
+# Copyright (c) 2003-2005 by Fredrik Lundh
+#
+# See the README file for information on usage and redistribution.
+#
+
+
+"""
+(Experimental) WCK-style drawing interface operations
+
+.. seealso:: :py:mod:`PIL.ImageDraw`
+"""
+from __future__ import annotations
+
+from typing import Any, AnyStr, BinaryIO
+
+from . import Image, ImageColor, ImageDraw, ImageFont, ImagePath
+from ._typing import Coords, StrOrBytesPath
+
+
+class Pen:
+    """Stores an outline color and width."""
+
+    def __init__(self, color: str, width: int = 1, opacity: int = 255) -> None:
+        self.color = ImageColor.getrgb(color)
+        self.width = width
+
+
+class Brush:
+    """Stores a fill color"""
+
+    def __init__(self, color: str, opacity: int = 255) -> None:
+        self.color = ImageColor.getrgb(color)
+
+
+class Font:
+    """Stores a TrueType font and color"""
+
+    def __init__(
+        self, color: str, file: StrOrBytesPath | BinaryIO, size: float = 12
+    ) -> None:
+        # FIXME: add support for bitmap fonts
+        self.color = ImageColor.getrgb(color)
+        self.font = ImageFont.truetype(file, size)
+
+
+class Draw:
+    """
+    (Experimental) WCK-style drawing interface
+    """
+
+    def __init__(
+        self,
+        image: Image.Image | str,
+        size: tuple[int, int] | list[int] | None = None,
+        color: float | tuple[float, ...] | str | None = None,
+    ) -> None:
+        if isinstance(image, str):
+            if size is None:
+                msg = "If image argument is mode string, size must be a list or tuple"
+                raise ValueError(msg)
+            image = Image.new(image, size, color)
+        self.draw = ImageDraw.Draw(image)
+        self.image = image
+        self.transform: tuple[float, float, float, float, float, float] | None = None
+
+    def flush(self) -> Image.Image:
+        return self.image
+
+    def render(
+        self,
+        op: str,
+        xy: Coords,
+        pen: Pen | Brush | None,
+        brush: Brush | Pen | None = None,
+        **kwargs: Any,
+    ) -> None:
+        # handle color arguments
+        outline = fill = None
+        width = 1
+        if isinstance(pen, Pen):
+            outline = pen.color
+            width = pen.width
+        elif isinstance(brush, Pen):
+            outline = brush.color
+            width = brush.width
+        if isinstance(brush, Brush):
+            fill = brush.color
+        elif isinstance(pen, Brush):
+            fill = pen.color
+        # handle transformation
+        if self.transform:
+            path = ImagePath.Path(xy)
+            path.transform(self.transform)
+            xy = path
+        # render the item
+        if op in ("arc", "line"):
+            kwargs.setdefault("fill", outline)
+        else:
+            kwargs.setdefault("fill", fill)
+            kwargs.setdefault("outline", outline)
+        if op == "line":
+            kwargs.setdefault("width", width)
+        getattr(self.draw, op)(xy, **kwargs)
+
+    def settransform(self, offset: tuple[float, float]) -> None:
+        """Sets a transformation offset."""
+        (xoffset, yoffset) = offset
+        self.transform = (1, 0, xoffset, 0, 1, yoffset)
+
+    def arc(
+        self,
+        xy: Coords,
+        pen: Pen | Brush | None,
+        start: float,
+        end: float,
+        *options: Any,
+    ) -> None:
+        """
+        Draws an arc (a portion of a circle outline) between the start and end
+        angles, inside the given bounding box.
+
+        .. seealso:: :py:meth:`PIL.ImageDraw.ImageDraw.arc`
+        """
+        self.render("arc", xy, pen, *options, start=start, end=end)
+
+    def chord(
+        self,
+        xy: Coords,
+        pen: Pen | Brush | None,
+        start: float,
+        end: float,
+        *options: Any,
+    ) -> None:
+        """
+        Same as :py:meth:`~PIL.ImageDraw2.Draw.arc`, but connects the end points
+        with a straight line.
+
+        .. seealso:: :py:meth:`PIL.ImageDraw.ImageDraw.chord`
+        """
+        self.render("chord", xy, pen, *options, start=start, end=end)
+
+    def ellipse(self, xy: Coords, pen: Pen | Brush | None, *options: Any) -> None:
+        """
+        Draws an ellipse inside the given bounding box.
+
+        .. seealso:: :py:meth:`PIL.ImageDraw.ImageDraw.ellipse`
+        """
+        self.render("ellipse", xy, pen, *options)
+
+    def line(self, xy: Coords, pen: Pen | Brush | None, *options: Any) -> None:
+        """
+        Draws a line between the coordinates in the ``xy`` list.
+
+        .. seealso:: :py:meth:`PIL.ImageDraw.ImageDraw.line`
+        """
+        self.render("line", xy, pen, *options)
+
+    def pieslice(
+        self,
+        xy: Coords,
+        pen: Pen | Brush | None,
+        start: float,
+        end: float,
+        *options: Any,
+    ) -> None:
+        """
+        Same as arc, but also draws straight lines between the end points and the
+        center of the bounding box.
+
+        .. seealso:: :py:meth:`PIL.ImageDraw.ImageDraw.pieslice`
+        """
+        self.render("pieslice", xy, pen, *options, start=start, end=end)
+
+    def polygon(self, xy: Coords, pen: Pen | Brush | None, *options: Any) -> None:
+        """
+        Draws a polygon.
+
+        The polygon outline consists of straight lines between the given
+        coordinates, plus a straight line between the last and the first
+        coordinate.
+
+
+        .. seealso:: :py:meth:`PIL.ImageDraw.ImageDraw.polygon`
+        """
+        self.render("polygon", xy, pen, *options)
+
+    def rectangle(self, xy: Coords, pen: Pen | Brush | None, *options: Any) -> None:
+        """
+        Draws a rectangle.
+
+        .. seealso:: :py:meth:`PIL.ImageDraw.ImageDraw.rectangle`
+        """
+        self.render("rectangle", xy, pen, *options)
+
+    def text(self, xy: tuple[float, float], text: AnyStr, font: Font) -> None:
+        """
+        Draws the string at the given position.
+
+        .. seealso:: :py:meth:`PIL.ImageDraw.ImageDraw.text`
+        """
+        if self.transform:
+            path = ImagePath.Path(xy)
+            path.transform(self.transform)
+            xy = path
+        self.draw.text(xy, text, font=font.font, fill=font.color)
+
+    def textbbox(
+        self, xy: tuple[float, float], text: AnyStr, font: Font
+    ) -> tuple[float, float, float, float]:
+        """
+        Returns bounding box (in pixels) of given text.
+
+        :return: ``(left, top, right, bottom)`` bounding box
+
+        .. seealso:: :py:meth:`PIL.ImageDraw.ImageDraw.textbbox`
+        """
+        if self.transform:
+            path = ImagePath.Path(xy)
+            path.transform(self.transform)
+            xy = path
+        return self.draw.textbbox(xy, text, font=font.font)
+
+    def textlength(self, text: AnyStr, font: Font) -> float:
+        """
+        Returns length (in pixels) of given text.
+        This is the amount by which following text should be offset.
+
+        .. seealso:: :py:meth:`PIL.ImageDraw.ImageDraw.textlength`
+        """
+        return self.draw.textlength(text, font=font.font)