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

[project]
name = "shortlist"
description = "Private recommendation rows for every user on your Plex server, with optional AI"
readme = "README.md"
license = "MIT"
# A genuine floor (3.12 syntax/stdlib is used), NOT a support claim: the shipped
# artifact is one Docker image built FROM python:3.12-slim, and CI tests only 3.12.
# Deliberately left open-ended rather than capped at <3.13 — a hard cap makes
# `pip install -e .` fail outright for contributors on a newer interpreter, and the
# code runs fine on them. The classifiers below state what's actually tested.
requires-python = ">=3.12"
dynamic = ["version"]
authors = [{ name = "Steven Adams" }]
keywords = ["plex", "plex-recommendations", "per-user", "recommendations", "recommendation-engine", "self-hosted", "selfhosted", "docker", "media-server", "plex-collections", "watch-history", "llm", "tmdb"]
classifiers = [
    "Development Status :: 5 - Production/Stable",
    "License :: OSI Approved :: MIT License",
    "Programming Language :: Python :: 3.12",
]
dependencies = [
    "plexapi>=4.17",
    "httpx>=0.27",
    "loguru>=0.7",
    "pydantic>=2.7",
    "fastapi>=0.115",
    "uvicorn>=0.30",
    "sqlalchemy>=2.0",
    "alembic>=1.13",
    "apscheduler>=3.10,<4",
    "itsdangerous>=2.1",
    "cryptography>=42.0",
    # Never imported by name: FastAPI requires it at request time to parse Form and UploadFile bodies
    # (poster upload), so dependency checkers report it as unused. Removing it breaks those endpoints.
    "python-multipart>=0.0.9",
    # Imported directly (requests/urllib3 in plex_pms and plex_reachability, starlette across the server,
    # anyio in the assistant stdio bridge) but otherwise only present because plexapi, fastapi and mcp pull them in.
    "requests>=2.31",
    "urllib3>=2.0",
    "starlette>=0.37",
    "anyio>=4.0",
    "mcp==2.3.0",
    "Authlib==1.9.5",
    # The PMS notification socket (`watch_stream`). Already in the lock transitively via uvicorn's
    # standard extras, but declared here because we import it directly — a transitive dependency that
    # disappears when uvicorn changes its extras would break the only source of partial-watch data.
    # >=14, not >=13: `websockets.connect` only became the asyncio client in 14.0. On 13.x it is the
    # legacy client, which takes `extra_headers` rather than `additional_headers` and whose `recv()`
    # is not documented as cancellation-safe — and `_listen` races it against the stop event with `ensure_future` + `wait`.
    "websockets>=14",
]

[project.optional-dependencies]
anthropic = ["anthropic>=0.40"]
openai = ["openai>=1.40"]
google = ["google-genai>=1.0"]
posters = ["pillow>=10.0"]
dev = [
    "pytest>=8.0",
    "pytest-xdist>=3.5",
    "pytest-cov>=5.0",
    # Duration-balanced CI sharding (see ci.yml's test-python matrix). Splitting by test
    # COUNT instead would be badly skewed — measured 51% of the suite's work in one third.
    "pytest-split>=0.9",
    "hypothesis>=6.100",
    "ruff>=0.5",
    "respx>=0.21",
    "playwright>=1.45",
    "pytest-rerunfailures>=14.0",
    "pytest-timeout>=2.3",
    "pillow>=10.0",  # so the poster text-renderer + upload-normalization tests run in CI

    # Local commit guard: runs the same ruff check + ruff format CI enforces, so a format slip is
    # caught at `git commit`, not in GitHub. Activate once per clone with `pre-commit install`.
    "pre-commit>=3.7",
    # scripts/build_llms_full.py and build_feed.py read the docs' YAML data files.
    "pyyaml>=6.0",
]

[project.urls]
Homepage = "https://github.com/stevezau/shortlist"
Issues = "https://github.com/stevezau/shortlist/issues"

[project.scripts]
shortlist-mcp-stdio = "shortlist.server.assistant.stdio_bridge:main"

[tool.hatch.version]
path = "shortlist/__init__.py"

[tool.hatch.build.targets.wheel]
packages = ["shortlist"]

[tool.ruff]
line-length = 120
target-version = "py312"

[tool.ruff.lint]
select = ["E", "F", "W", "I", "UP", "B", "SIM", "RUF"]

[tool.ruff.lint.isort]
known-first-party = ["shortlist"]

[tool.pytest.ini_options]
testpaths = ["tests"]
# tests.scratch enforces disk-backed storage and these bounds, including CLI overrides.
tmp_path_retention_count = 2
tmp_path_retention_policy = "failed"
# e2e needs a built SPA + a browser; it runs in its own CI job via `pytest -m e2e`.
# Keep ordinary runs bounded on shared hosts, regardless of the host's CPU count.
addopts = "-n 2 -m 'not e2e' --cov=shortlist --cov-report=term-missing"
# A hung test once spun a worker at 100% CPU for 40 minutes and left the run on one worker, never to
# finish. `thread` dumps every thread's stack and ends that worker, so a hang fails one test and shows
# where it was stuck; xdist starts a replacement worker.
timeout = 300
timeout_method = "thread"
markers = [
    "integration: crosses module boundaries",
    "plex: requires a real Plex server (skipped in CI)",
    "slow: long-running",
    "real_migrations: needs an empty config dir, not the pre-seeded schema template",
    "e2e: Playwright vs an in-process app (uvicorn + built SPA) + fake_plex",
]

[tool.coverage.report]
fail_under = 80