Run any Skill in Manus with one click

test-with-replay-cache

Stars1

Forks0

UpdatedJune 7, 2026 at 07:54

Writes and debugs python-manta tests that parse real Dota replays without re-parsing, using the module-scoped cached-fixture system (conftest.py + caching_parser.py), the GCS replay auto-download, the correct pytest markers, and the right subset commands. Use whenever the user wants to write or fix a test for the parser, add a fixture, debug a test that can't find a replay, speed up slow/re-parsing tests, add a unit vs integration test, deal with the coverage gate, or configure DOTA_REPLAY_CACHE. Use this even when the user just says "test the wards collector" or "the tests are re-parsing".

Installation

Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.

Run Skill in Manus

Source

DeepBlueCoding

DeepBlueCoding/python-manta

View GitHub Repository View Creator Repositories

Download

Run Skill in Manus

Related occupationsSOC

Based on SOC occupation classification

Software Quality Assurance Analysts and TestersComputer and Mathematical Occupations·SOC 15-1253

SKILL.md

readonly

name

test-with-replay-cache

description

Testing with the replay cache

Parsing a full replay takes 5+ minutes, so tests share parsed results via a global cache and module-scoped fixtures. Read CLAUDE.md ("Test Caching System") first — this skill adds the locations, cache-key behavior, and exact run commands.

HARD CONSTRAINT: a full test run may already be in progress on this machine. Prefer reading code and running narrow subsets over launching the whole suite.

Architecture

tests/caching_parser.py — wraps the real Parser with global caches. parse() is keyed on (demo_path, json.dumps(kwargs, sort_keys=True)); build_index() on (demo_path, interval_ticks); snapshot() on (demo_path, target_tick, include_illusions). IDENTICAL parse() kwargs reuse the cache; DIFFERENT kwargs create a new (expensive) parse. So reuse an existing fixture's kwargs EXACTLY when you can.
tests/conftest.py — defines ALL fixtures, every one scope="module". Examples: parser, parser_secondary, combat_log, attacks_result, melee_attacks, ranged_attacks, wards_result (kwargs parser.parse(wards={})), attacks_result (kwargs attacks={"max_events": 0}), snapshot_30k (tick 30000), demo_index (interval_ticks 1800).

Two hard rules (banned in CLAUDE.md)

NEVER define a local def parser() (or any duplicate) fixture in a test file — use the conftest one.
NEVER from python_manta import Parser in a test file. Get the parser only via the parser fixture argument. To parse with custom options inside a test, call parser.parse(...) on the injected fixture (still cached).

Need a reusable result? Add a scope="module" fixture to tests/conftest.py (never in the test file), reusing existing kwargs where possible:

@pytest.fixture(scope="module")
def foo_result(parser):
    """Cached foo parsing result."""
    result = parser.parse(foo={})
    assert result.success, f"Failed to parse foo: {result.error}"
    return result.foo

Replay location / download

Replays auto-download from GCS https://storage.googleapis.com/dota-replays/<match_id>.dem via tests/replay_cache.py. Cache dir resolution order: DOTA_REPLAY_CACHE env var → .data/replays/ → ~/.cache/dota-replays/ → project replays/. Point at a local copy to avoid re-downloading:

DOTA_REPLAY_CACHE=/path/to/replays uv run pytest ...

Primary match id is 8447659831 (Team Spirit vs Tundra); secondary is 8461956309. A "can't find replay" failure is almost always cache-dir / network, not a code bug.

Layout, markers, and what CI runs

tests/python_manta/** — unit-ish; this is the suite CI runs against built wheels (pytest tests/python_manta/ --no-cov). Mark units -m unit.
tests/scenarios/** — integration/slow; existing files start with pytestmark = [pytest.mark.slow, pytest.mark.integration].
Also: tests/parity, tests/cli, tests/performance.
pytest.ini declares ONLY unit and integration (not slow) and sets --strict-markers. The scenario files use slow regardless, so it must be registered at run time (e.g. by a plugin) or a --strict-markers run errors on it. Prefer selecting by integration for scenario tests; if you add a new marker, declare it under markers = in pytest.ini first.

Golden values, not type checks

Assert REAL expected values from the actual replay. tests/scenarios/test_wards.py is the model: EXPECTED_TOTAL_WARDS == 105, EXPECTED_OBSERVER_WARDS == 38, EXPECTED_SENTRY_WARDS == 67. Do NOT write isinstance / len >= 0 / field-existence checks.

Running subsets (and the coverage caveat)

pytest.ini enforces --cov-fail-under=90 globally; the 90% gate only passes on the FULL suite, so a partial run reports a coverage FAILURE. Use --no-cov while iterating:

cd /home/juanma/projects/python-manta
uv run pytest tests/scenarios/test_wards.py --no-cov            # one scenario file
uv run pytest tests/python_manta/enums/test_enums.py --no-cov   # one unit file
uv run pytest tests/python_manta/ -m unit --no-cov              # all units, no demos needed

Helper runner (does coverage/markers for you): uv run python run_tests.py --unit / --integration / --all --coverage.

More from this repository

same repository

add-collector

DeepBlueCoding/python-manta

Adds a new single-pass data collector to python-manta end-to-end (Go CGO layer, Python Pydantic models, parse() kwarg, fixture, scenario test, docs) so it becomes reachable as parser.parse(<name>={...}). Use whenever the user wants to add a new collector, track or capture a new kind of event/entity during parsing (wards-style, attacks-style), expose new dotabuff/manta data through parse(), add a parse() collector kwarg, or create a new *Result/*Config result type. Use this even if the user only describes the data they want surfaced (e.g. "track every tower attack", "capture courier deaths") without saying the word "collector".

2026-06-071

add-or-fix-enum-model

DeepBlueCoding/python-manta

Adds or modifies the Pydantic models and game-data enums in python-manta's manta_python.py (heroes, items, neutral items, rune/entity/combat-log/damage types) following the project's strict canonical-enum-plus-alias philosophy. Use whenever the user wants to add a new hero/item/enum, add an alias for an internal replay name, add or change a Pydantic model field, add a display_name, resolve npc_dota_hero_* / item_* names, or update enums for a new Dota patch. Use this even when the user just says "map this name", "add this hero", or "the enum is missing X".

2026-06-071

release-version

DeepBlueCoding/python-manta

Cuts a python-manta release (dev build to TestPyPI or final to PyPI) and/or bumps the locked upstream dotabuff/manta version, using the correct PEP 440 4-part version scheme and the exact git tag format the GitHub Actions CI keys on. Use whenever the user wants to release a new version, bump the version, publish to PyPI/TestPyPI, cut a dev/prerelease, tag a release, upgrade or sync the manta version, or change .manta-version. Use this even if the user just says "ship it", "make a dev build", or "update manta to vX.Y.Z".

2026-06-071

write-docs

DeepBlueCoding/python-manta

Authors and updates the python-manta mkdocs documentation site (guides, API reference, reference pages) consistent with the mkdocs.yml nav, mike versioning, llms.txt generation, and the project's time-formatting and data-extraction-only content rules. Use whenever the user wants to write or update docs, add a guide, document a new feature/collector in the docs site, update the API reference, add a reference page, edit mkdocs, or work on the GitHub Pages / llms.txt site. Use this even when the user just says "document this" or "add it to the docs".

2026-06-071

name

test-with-replay-cache

description

Testing with the replay cache

HARD CONSTRAINT: a full test run may already be in progress on this machine. Prefer reading code and running narrow subsets over launching the whole suite.

Architecture

tests/caching_parser.py — wraps the real Parser with global caches. parse() is keyed on (demo_path, json.dumps(kwargs, sort_keys=True)); build_index() on (demo_path, interval_ticks); snapshot() on (demo_path, target_tick, include_illusions). IDENTICAL parse() kwargs reuse the cache; DIFFERENT kwargs create a new (expensive) parse. So reuse an existing fixture's kwargs EXACTLY when you can.
tests/conftest.py — defines ALL fixtures, every one scope="module". Examples: parser, parser_secondary, combat_log, attacks_result, melee_attacks, ranged_attacks, wards_result (kwargs parser.parse(wards={})), attacks_result (kwargs attacks={"max_events": 0}), snapshot_30k (tick 30000), demo_index (interval_ticks 1800).

Two hard rules (banned in CLAUDE.md)

NEVER define a local def parser() (or any duplicate) fixture in a test file — use the conftest one.
NEVER from python_manta import Parser in a test file. Get the parser only via the parser fixture argument. To parse with custom options inside a test, call parser.parse(...) on the injected fixture (still cached).

Need a reusable result? Add a scope="module" fixture to tests/conftest.py (never in the test file), reusing existing kwargs where possible:

@pytest.fixture(scope="module")
def foo_result(parser):
    """Cached foo parsing result."""
    result = parser.parse(foo={})
    assert result.success, f"Failed to parse foo: {result.error}"
    return result.foo

Replay location / download

DOTA_REPLAY_CACHE=/path/to/replays uv run pytest ...

Primary match id is 8447659831 (Team Spirit vs Tundra); secondary is 8461956309. A "can't find replay" failure is almost always cache-dir / network, not a code bug.

Layout, markers, and what CI runs

tests/python_manta/** — unit-ish; this is the suite CI runs against built wheels (pytest tests/python_manta/ --no-cov). Mark units -m unit.
tests/scenarios/** — integration/slow; existing files start with pytestmark = [pytest.mark.slow, pytest.mark.integration].
Also: tests/parity, tests/cli, tests/performance.
pytest.ini declares ONLY unit and integration (not slow) and sets --strict-markers. The scenario files use slow regardless, so it must be registered at run time (e.g. by a plugin) or a --strict-markers run errors on it. Prefer selecting by integration for scenario tests; if you add a new marker, declare it under markers = in pytest.ini first.

Golden values, not type checks

Running subsets (and the coverage caveat)

pytest.ini enforces --cov-fail-under=90 globally; the 90% gate only passes on the FULL suite, so a partial run reports a coverage FAILURE. Use --no-cov while iterating:

cd /home/juanma/projects/python-manta
uv run pytest tests/scenarios/test_wards.py --no-cov            # one scenario file
uv run pytest tests/python_manta/enums/test_enums.py --no-cov   # one unit file
uv run pytest tests/python_manta/ -m unit --no-cov              # all units, no demos needed

Helper runner (does coverage/markers for you): uv run python run_tests.py --unit / --integration / --all --coverage.