pyproject.toml

[build-system]
requires = ["setuptools>=68", "wheel"]
build-backend = "setuptools.build_meta"

[project]
name = "marathon-ocr-core"
version = "0.7.0"
description = "Shared OCR engine — Visual Evidence Charter + Arabic corrections + unified command system"
requires-python = ">=3.10"
license = {text = "MIT"}
dependencies = ["pyyaml>=6.0"]

[project.scripts]
ocr-core = "ocr_core.commands.adapters.cli:main"

[project.optional-dependencies]
tesseract = ["pytesseract>=0.3.10", "pillow>=10.0"]
paddle = ["paddleocr>=2.7", "pillow>=10.0", "numpy>=1.24"]
surya = ["surya-ocr>=0.6"]
easyocr = ["easyocr>=1.7", "pillow>=10.0", "numpy>=1.24"]
preprocess = ["opencv-python-headless>=4.8", "pillow>=10.0", "numpy>=1.24", "imagehash>=4.3", "rapidfuzz>=3.6"]
benchmarks = ["rapidfuzz>=3.6", "pyyaml>=6.0", "click>=8.1", "rich>=13.0"]
golden = ["arabic-reshaper>=3.0", "python-bidi>=0.4.2", "pillow>=10.0", "numpy>=1.24"]
# كشف الكلمات في الخط اليدوي (المرحلة 1) — الكاشف الدقيق ONNX
# (يرث [preprocess] لأن حزمة ocr_core.preprocess تستورد cv2 في __init__؛
#  كاشفا projection/heuristic نفسها لا تحتاج سوى numpy+pillow)
word-detect = ["marathon-ocr-core[preprocess]", "onnxruntime>=1.17", "huggingface_hub>=0.20"]
# jina-ocr-v1 — محرك VLM للمستندات الكاملة (المرحلة 4)
# ⚠️ ترخيص الأوزان: CC BY-NC 4.0 (غير تجاري) — الكود هنا MIT،
#    الأوزان تُحمّل وقت التشغيل من HuggingFace.
jina = ["torch>=2.2", "transformers>=4.45", "torchvision>=0.17", "accelerate>=0.29", "safetensors>=0.4", "pillow>=10.0"]
# مزيج VLM الموصى به (jina + PDF→صور + تصدير DOCX لأوامر commands)
vlm = ["marathon-ocr-core[jina]", "pymupdf>=1.23", "python-docx>=1.1"]
pdf = ["pymupdf>=1.23"]
docx = ["python-docx>=1.1"]
# ⚠️ all لا يتضمن jina/surya/word-detect — تراخيص أوزان أو تبعيات ثقيلة
#    يختارها المستخدم صراحةً.
all = ["pytesseract>=0.3.10", "pillow>=10.0", "paddleocr>=2.7", "numpy>=1.24", "easyocr>=1.7", "opencv-python-headless>=4.8", "imagehash>=4.3", "rapidfuzz>=3.6", "click>=8.1", "rich>=13.0", "arabic-reshaper>=3.0", "python-bidi>=0.4.2", "pymupdf>=1.23", "python-docx>=1.1"]
dev = ["pytest>=8.0", "ruff>=0.6"]
ui = ["PySide6>=6.5"]
ui-dev = ["PySide6>=6.5", "pytest-qt>=4.4"]

[tool.setuptools.packages.find]
where = ["src"]

[tool.setuptools.package-data]
ocr_core = ["rules/data/*.yaml", "postprocess/data/*.jsonl", "postprocess/data/*.json", "config/*.yaml", "benchmarks/golden/fonts/*.ttf", "benchmarks/golden/fonts/*.txt"]