[build-system]
requires = ["hatchling"]
build-backend = "hatchling.build"

[project]
name = "pdfmux"
version = "1.8.7"
description = "Self-healing PDF extraction for RAG — #2 of 8 engines on opendataloader-bench (0.903). Flags pages it can't read instead of dropping them, and certifies any extractor's output for silent drops. Verify signed manifests offline, free. Per-page confidence, 7-tool MCP server, LangChain/LlamaIndex loaders. LlamaParse alternative."
readme = "README.md"
license = "MIT"
requires-python = ">=3.11"
authors = [
    { name = "Nameet Potnis" },
]
keywords = [
    "pdf",
    "pdf-to-markdown",
    "pdf-extraction",
    "rag",
    "llm",
    "langchain",
    "llamaindex",
    "mcp",
    "ocr",
    "confidence-scoring",
    "self-healing",
    "llamaparse-alternative",
    "document-ai",
    "pymupdf-alternative",
    "docling-alternative",
    "table-extraction",
    "structured-extraction",
    "ai-document-processing",
    "pdf-parser",
    "scanned-pdf",
    "document-ingestion",
    "rag-pipeline",
    "claude-desktop",
]
classifiers = [
    "Development Status :: 5 - Production/Stable",
    "Intended Audience :: Developers",
    "License :: OSI Approved :: MIT License",
    "Programming Language :: Python :: 3",
    "Programming Language :: Python :: 3.11",
    "Programming Language :: Python :: 3.12",
    "Programming Language :: Python :: 3.13",
    "Topic :: Text Processing :: Markup :: Markdown",
    "Topic :: Scientific/Engineering :: Artificial Intelligence",
    "Topic :: Office/Business",
    "Topic :: Software Development :: Libraries :: Python Modules",
    "Typing :: Typed",
    "Environment :: Console",
]
dependencies = [
    "typer>=0.9.0",
    # UPPER BOUNDS ARE LOAD-BEARING — do not relax without running
    # tests/test_dependency_regression.py against the new resolve.
    # pymupdf4llm renumbered 0.3.4 -> 1.28.0 to track pymupdf. On 1.28.0
    # to_markdown() returns headings only and drops body prose entirely:
    # eval/fixtures/good-digital-00.pdf yields 0 chars where 0.3.4 yields 218
    # (raw fitz sees 214). That silently guts every extraction pdfmux makes —
    # the exact failure mode this product exists to catch. Found 2026-07-20.
    "pymupdf>=1.24.0,<1.29",  # garbage filter handles 1.27 find_tables() regression
    "pymupdf4llm>=0.0.10,<1.0",
    "rich>=13.0.0",
    "python-bidi>=0.6.0",  # BiDi algorithm for Arabic/Hebrew RTL text reordering
    "cryptography>=42.0",  # offline Ed25519 verification of Cloud signed manifests (pdfmux verify-manifest)
]

[project.optional-dependencies]
# mcp LOWER bound is load-bearing: mcp 2.0.0 deleted mcp.server.fastmcp and
# replaced FastMCP with mcp.server.mcpserver.MCPServer. mcp_server.py and
# mcp_extract.py now import the 2.x path, which does not exist in 1.x — so
# >=2.0.0 is a floor, not a preference. Do NOT re-add an upper bound to dodge a
# future bump (that is what commit 6a4f78a did); port the import surface instead.
serve = ["mcp>=2.0.0", "uvicorn>=0.30.0"]
tables = ["docling>=2.0.0"]
opendataloader = ["opendataloader-pdf>=2.0.0"]
ocr = ["rapidocr>=3.0.0", "onnxruntime>=1.19.0"]
ocr-heavy = ["surya-ocr>=0.6.0"]
marker = ["marker-pdf>=1.0.0"]
llm = ["google-genai>=1.0.0"]
llm-claude = ["anthropic>=0.40.0"]
llm-openai = ["openai>=1.0.0"]
llm-ollama = ["ollama>=0.4.0"]
llm-mistral = ["mistralai>=2.7.1"]
llm-all = ["pdfmux[llm,llm-claude,llm-openai,llm-ollama,llm-mistral]"]
langchain = ["langchain-core>=0.2.0"]
llamaindex = ["llama-index-core>=0.10.0"]
watch = ["watchdog>=3.0.0"]
all = ["pdfmux[serve,tables,ocr,opendataloader,marker,llm-all,langchain,llamaindex,watch]"]
dev = [
    "pytest>=8.0.0",
    "pytest-asyncio>=0.23.0",
    "ruff>=0.15.22",
    # tests/test_profiles.py exercises profile save/load, which needs PyYAML.
    # It used to arrive transitively and the suite passed by luck; pinning the
    # PDF stack removed that accident. Declared here so the profiles feature is
    # actually covered rather than skipped by an environment quirk.
    "pyyaml>=6.0",
]

[project.scripts]
pdfmux = "pdfmux.cli:app"
pdfmux-extract = "pdfmux.mcp_extract:run_server"

[project.urls]
Homepage = "https://pdfmux.com"
Documentation = "https://github.com/NameetP/pdfmux#readme"
Repository = "https://github.com/NameetP/pdfmux"
Issues = "https://github.com/NameetP/pdfmux/issues"
Changelog = "https://github.com/NameetP/pdfmux/blob/main/CHANGELOG.md"

[tool.hatch.build.targets.wheel]
packages = ["src/pdfmux"]

[tool.ruff]
target-version = "py311"
line-length = 100

[tool.ruff.lint]
select = ["E", "F", "I", "N", "W", "UP"]

[tool.ruff.lint.per-file-ignores]
"tests/**" = ["E501", "F401", "I001", "N818"]
"src/pdfmux/mcp_server.py" = ["E501"]  # MCP tool docstrings are user-facing
"src/pdfmux/cli.py" = ["E501"]  # CLI help text is user-facing
"src/pdfmux/headings.py" = ["E501"]  # regex patterns for date matching
"src/pdfmux/image_table_ocr.py" = ["E741"]  # short loop variables
"eval/ab_models.py" = ["E501"]  # report tables + example commands are presentation strings

[tool.pytest.ini_options]
testpaths = ["tests"]
asyncio_mode = "auto"