pytest.ini

[pytest]
# AutoBot pytest configuration

# Ensure backend and shared modules are importable before collection starts.
# pytest processes `pythonpath` before any test collection, so bare imports
# like `from models.task_context import ...` resolve correctly.
# Requires pytest >= 7.0 (built-in mechanism; more reliable than conftest.py
# sys.path manipulation which runs too late to prevent collection errors).
# Issue: #1884
# pipeline-scripts (#13880): its tests do bare imports of their subject
# (`import audit_unwired_trackers`), and --import-mode=importlib below does NOT
# put a test's own directory on sys.path. Collecting that directory without
# this entry fails every shard at import time.
# .claude/skills/claims-audit (#14986): its four suites import their subjects
# bare (`from verifiers import ...`, `from file_issues import ...`). The
# directory cannot become an importable package path -- neither `.claude` nor
# `claims-audit` is a Python identifier -- so #13086's pkgutil.extend_path
# route does not transfer, and --import-mode=importlib does not put a test's
# own directory on sys.path. This is the same entry `pipeline-scripts` gets
# above and for the same reason. It replaces three sys.path.insert calls that
# resolved only because an earlier-collected sibling ran first.
#
# It is declared BEFORE the SDK entry below on purpose: that one is required
# to stay last (see its note), and this one has no ordering requirement of its
# own -- claims-audit publishes `verifiers`, `file_issues`, `generate_report`
# and a JSON-only `patterns/`, none of which shadows anything (#14986).
# libs/autobot-sdk-python (#13286): `libs/autobot-sdk-python/tests/test_integration.py`
# carries `pytestmark = pytest.mark.integration` and was selected by no workflow until
# marker-tests.yml gained that root. Its first run reported four
# `ModuleNotFoundError: No module named 'autobot_sdk'` — the SDK package sits under
# `libs/autobot-sdk-python/`, which nothing put on the path. That is a packaging gap,
# not a product one, so it is fixed here rather than marked around. The top-level name
# `autobot_sdk` is unique in this repository, so that name collides with nothing.
#
# ORDER IS LOAD-BEARING, and this entry must stay LAST. `libs/autobot-sdk-python`
# also holds a `tests/__init__.py`, so it publishes a top-level `tests` package,
# and five backend e2e modules do `from tests.test_helpers import ...` expecting
# `autobot-backend/tests`. pytest inserts these entries in declared order, so the
# backend wins as written; moving this entry earlier would repoint those imports
# silently, and the error would name the helper rather than this line.
# repo_tests/marker_suite_root_coverage_test.py fails if the order changes.
pythonpath = . autobot-backend autobot_shared pipeline-scripts .claude/skills/claims-audit libs/autobot-sdk-python

# Test markers
markers =
    integration: Integration tests that may require external services
    requires_backend: Tests written against a running backend API (autobot-infrastructure harness)
    slow: Tests that take a long time to run
    distributed: Tests for distributed multi-VM scenarios
    performance: Performance and load tests
    security: Security-related tests
    real_kb: Tests that run against a real (in-memory) ChromaDB instance
    migration_gate: Alembic migration-path tests requiring a disposable Postgres (#10001/#10026)
    xdist_group: pytest-xdist group marker — tests with the same group run on the same worker
    e2e: End-to-end tests exercising a full service stack (#13549)
    high_priority: Release-blocking security and test-coverage assertions (#13549)
    reach_floor: Reach floors checked against the LIVE tree — one full sweep per declaration, so
        pre-push deselects them and CI runs them (#17810). The empty-tree proof is NOT marked.
    llm_judge: LLM-judged quality assertions — skipped unless AUTOBOT_LLM_JUDGE_TESTS=1 and a provider is configured (#11521)

# Test discovery - colocated tests live next to source files (#734)
# #12501: added autobot-tts-worker — its tests/ mirror tts-worker.py.j2
# template logic in isolation (no torch/pocket_tts needed) and were
# previously never collected by CI.
#
# #13084: autobot-backend/ and autobot-slm-backend/ each define top-level
# packages with identical names (api, services, user_management, middleware,
# migrations). Running ALL of testpaths in ONE pytest invocation makes
# whichever backend collects first bind those dotted names in sys.modules for
# the rest of the session, breaking the other backend's own tests. A full
# local/CI run MUST use two separate invocations instead of a single bare
# `pytest`: one covering `autobot-backend autobot_shared autobot-tts-worker
# repo_tests`, a second covering `autobot-slm-backend` alone (see
# .github/workflows/ci.yml's "Run unit tests with coverage gate" step).
# Running a single test file/directory under either backend is unaffected.
#
testpaths =
    autobot-backend
    autobot-slm-backend
    autobot-tts-worker
    autobot_shared
    repo_tests
    # #13368: `tools/` holds 215 passing tests (lint checkers, codemods) that
    # nothing ran — it was absent here, so CI collected none of them and a
    # broken one could not be noticed. Its own tests import via the package
    # path (`tools.lint.*`, `tools.codemods.*`), which resolves because both
    # `tools/__init__.py` files call pkgutil.extend_path (#13086).
    tools
    # #13879: ci.yml has passed `scripts` to pytest for a while, but testpaths
    # never listed it — so a bare local `pytest` silently skipped every test in
    # there. The two lists had drifted, which is how a red test hides.
    scripts
    # #13880: same defect as #13368 one directory over — `pipeline-scripts/`
    # holds tests for the CI guard wrappers and nothing collected them, so a
    # test that had been red for as long as the bug it covered went unnoticed.
    # A guard's own test must run, or the guard is unverified twice over.
    pipeline-scripts
    # #14884: the same defect a third time, and the sharpest case of it. The 18
    # suites under autobot-infrastructure/shared/scripts/hooks/ are the ONLY
    # tests of the pre-commit hooks themselves, and nothing collected them: not
    # this list, not any of the three CI invocations, and not
    # autobot-infrastructure/shared/tests/pytest.ini (whose `testpaths = unit
    # integration e2e` does not reach shared/scripts/). So `git worktree list`
    # failing closed, and the supply-chain check on tag-pinned third-party
    # actions, were both guarded by tests that ran nowhere. 159 tests, measured
    # green before wiring — this is a wiring fix, not a quarantine.
    #
    # NARROW ON PURPOSE. `autobot-infrastructure` as a whole does NOT collect:
    # 23 collection errors when this was measured. The second reason recorded
    # here -- shared/scripts/test_alertmanager.py calling sys.exit(0) at import
    # and killing pytest with an INTERNALERROR -- is SPENT: #14917 put it behind
    # a main() guard (see its lines 15 and 67-68). Widening is #15476 (#15178).
    autobot-infrastructure/shared/scripts/hooks
    # #14986: the four suites here hold 42 test functions CI had never run,
    # because this list is an allowlist and `.claude` was not on it. The skill
    # is repo tooling whose findings are used in review, so the #13368 argument
    # for `tools/` applies unchanged -- and measuring it proved the point: 6 of
    # the 42 were red, against real defects in the verifiers (a truncated
    # endpoint path, a subject extractor that returned "the", two config
    # phrasings that reached no verifier, an infrastructure category no
    # verifier claimed, and a citation stripped down to a basename). All six
    # are fixed and all 42 pass.
    #
    # NARROW ON PURPOSE, like the hooks entry above. `.claude/` as a whole does
    # NOT collect: it is harness territory, and on a working checkout it holds
    # agent scratch that is not always parseable Python. The imports resolve
    # through this file's `pythonpath` entry, not through a sys.path shim in
    # each module, and `verifiers/test_verifier.py` -- test-shaped by
    # `python_files` but an implementation -- declares `__test__ = False`
    # rather than being renamed out of its package's `<kind>_verifier.py`
    # convention.
    .claude/skills/claims-audit
    # #15051: `autobot-infrastructure/shared/tests` and `libs` were named by NO
    # pytest invocation in any workflow and by no entry here, so a bare local
    # `pytest` skipped 28 tests in silence -- 12 of them marker-carrying (later
    # picked up by marker-tests.yml's third invocation, #13286) and 17 selected
    # by nothing at all, ever, not even the marker suite. Re-measured on this
    # branch: 11 of the original 28 gained a module-level `pytestmark` since
    # (#14979/#15166), leaving 17 (13 in test_architecture_compliance.py, 4 in
    # test_redis_db_ssot.py) that ran nowhere until this entry.
    #
    # Both roots named together, like the hooks/claims-audit pairs above: `libs`
    # carries `libs/autobot-sdk-python/tests`, whose `test_integration.py` needs
    # `libs/autobot-sdk-python` on `pythonpath` (already last on that list, see
    # above) to import `autobot_sdk` at all -- and #15161 found that pytest's
    # rootdir resolution only lands on the repository root's own pytest.ini
    # when `autobot-infrastructure/shared/tests` is invoked ALONGSIDE a sibling
    # path outside it, which every root already on this list provides.
    autobot-infrastructure/shared/tests
    libs
# #7082: dropped `*.e2e_test.py` / `*.integration_test.py` (dotted) — they
# break pytest's path-to-module conversion. Files renamed to `*_e2e_test.py`
# / `*_integration_test.py` / `*_performance_test.py` (underscore form),
# which all fall under the `*_test.py` pattern below.
python_files = test_*.py *_test.py
python_classes = Test*
python_functions = test_*

# Exclude directories that are not real test code:
#   - autobot-slm-backend/ansible/ contains Ansible-synced node copies with duplicate tests
#   - autobot-npu-worker/resources/ contains Windows-only tests requiring PySide6
#   - Standard tooling/cache directories
norecursedirs =
    .git
    .tox
    .venv
    venv
    __pycache__
    node_modules
    .worktrees
    dist
    build
    *.egg-info

# Output options
addopts =
    --verbose
    --strict-markers
    --tb=short
    --asyncio-mode=auto
    --import-mode=importlib
    --ignore=autobot-slm-backend/ansible
    --ignore=autobot-npu-worker/resources
    # pytest-xdist parallelism (-n auto --dist loadscope) is applied per
    # full-suite invocation (the ci.yml coverage gate), NOT globally here:
    # several CI jobs (e.g. startup-import-smoke) install bare pytest without
    # pytest-xdist and run from this rootdir, so a global -n in addopts would
    # error them with "unrecognized arguments: -n" and break a required gate.
    # Proven stable with --dist loadscope (#10398).

# Coverage options
# Run with: pytest --cov=autobot-backend --cov=autobot-slm-backend --cov-report=html
#
# The [coverage:run] / [coverage:report] sections that used to live here have
# MOVED to .coveragerc (#10691). They were inert in this file: coverage.py reads
# only .coveragerc, .coveragerc.toml, setup.cfg, tox.ini and pyproject.toml —
# never pytest.ini — so `omit` and `fail_under` here never applied. Do not add
# them back here; edit .coveragerc.