gabriel / musehub public
test_intel_bounded_providers.py python
88 lines 4.5 KB
Raw
sha256:b4b999434d042d74edc7032f1f2cfd20fda5b3580fe85966351db5b8eef0abb3 feat(intel): every grammar on the hub, and every intel job … Opus 5.5 minor ⚠ breaking 1 day ago
1 """Bounded cost and honest truncation for the history-mining providers — musehub#8 release gate (residual cost hotspots).
2
3 * Coupling counted every file pair of every commit into one dict (memory O(commits x files^2)); it now uses the bounded
4 ``muse.plugins.code.analysis.coupled_files`` (shared with ``muse code coupling``), as entangle does.
5 * The 10,000-commit walk cap of coupling / entangle / velocity was silent: ``truncated`` described the pairs, not the window.
6
7 Run ONLY this file, against a private Postgres (never the live hub database)::
8
9 TEST_DATABASE_URL=postgresql+asyncpg://musehub:musehub@<private-host>:<port>/musehub_test python -P -m pytest tests/test_intel_bounded_providers.py
10
11 H1 The bounded coupling algorithm returns exactly what exhaustive counting returned (pairs, order, tie-breaks, truncation), on randomised histories.
12 H2 It holds no more than a few multiples of ``max_pairs`` candidate pairs at any time, however many files and commits there are.
13 H3 A walk that hit the cap says so (``walk_truncated``) in coupling, entangle and velocity; one that did not, says it did not.
14 """
15 from __future__ import annotations
16
17 import random
18 from collections import defaultdict
19
20 import pytest
21 import pytest_asyncio
22
23 from muse.plugins.code.analysis import coupled_files
24 from musehub.services import musehub_intel_providers as P
25 from tests.factories import create_repo
26 from tests.test_intel_snapshot_providers import World
27
28
29 @pytest_asyncio.fixture
30 async def session_factory(db_session, session_factory): # noqa: F811
31 return session_factory
32
33
34 def exhaustive(commit_files: dict[str, set[str]], *, max_files: int, min_co: int, max_pairs: int) -> list[tuple[str, str, int]]:
35 counts: dict[tuple[str, str], int] = defaultdict(int)
36 for files in commit_files.values():
37 if len(files) > max_files:
38 continue
39 ordered = sorted(files)
40 for i in range(len(ordered)):
41 for j in range(i + 1, len(ordered)):
42 counts[(ordered[i], ordered[j])] += 1
43 pairs = sorted(((a, b, n) for (a, b), n in counts.items() if n >= min_co), key=lambda p: (-p[2], p[0], p[1]))
44 return pairs[:max_pairs]
45
46
47 def random_history(rng: random.Random) -> dict[str, set[str]]:
48 files = [f"f{i:02d}.py" for i in range(rng.randint(3, 40))]
49 return {f"c{i}": set(rng.sample(files, rng.randint(1, min(len(files), 12)))) for i in range(rng.randint(1, 60))}
50
51
52 @pytest.mark.parametrize("seed", range(60))
53 def test_h1_the_bounded_algorithm_equals_exhaustive_counting(seed: int) -> None:
54 rng = random.Random(seed)
55 history = random_history(rng)
56 for max_pairs in (1, 3, 200):
57 want = exhaustive(history, max_files=8, min_co=2, max_pairs=max_pairs)
58 got = coupled_files(history.values(), max_files_per_commit=8, min_co_changes=2, limit=max_pairs)
59 assert got == want, (seed, max_pairs)
60
61
62 def test_h2_candidate_memory_is_bounded_by_max_pairs_not_by_the_history() -> None:
63 rng = random.Random(7)
64 files = [f"m{i:03d}.py" for i in range(150)]
65 history = {f"c{i}": set(rng.sample(files, 60)) for i in range(120)} # ~1,770 pairs per commit, ~200k distinct pairs
66 stats = {"peak": 0}
67 pairs = coupled_files(history.values(), max_files_per_commit=200, min_co_changes=2, limit=20, stats=stats)
68 assert len(pairs) == 20
69 assert stats["peak"] <= 5 * 20 + 150 # 4x the cut plus one file's worth of fresh candidates; the exhaustive dict held ~10,000+
70 assert pairs == exhaustive(history, max_files=200, min_co=2, max_pairs=20)
71
72
73 @pytest.mark.asyncio
74 @pytest.mark.parametrize("cls,key", [(P.CouplingProvider, "intel.code.coupling"), (P.EntangleProvider, "intel.code.entangle"), (P.VelocityProvider, "intel.code.velocity")])
75 async def test_h3_a_walk_that_hit_the_cap_says_so(session_factory, monkeypatch, cls, key) -> None:
76 async with session_factory() as s:
77 repo = await create_repo(s, owner="gabriel", slug=f"walk-{key.rsplit('.', 1)[1]}")
78 w = World(s, repo.repo_id)
79 head = None
80 for i in range(5):
81 head = await w.commit(f"c{i}", {"m.py": f"x = {i}\n"}, [f"c{i - 1}"] if i else None, minute=i)
82 await s.commit()
83 monkeypatch.setattr(cls, "_MAX_WALK", 3)
84 cut = (await cls().compute(s, w.repo_id, head, {}))[0]
85 assert cut[0] == key and cut[1]["walk_truncated"] is True and cut[1]["commits_analysed"] == 3
86 monkeypatch.setattr(cls, "_MAX_WALK", 5) # exactly the history length: nothing was left out
87 whole = (await cls().compute(s, w.repo_id, head, {}))[0]
88 assert whole[1]["walk_truncated"] is False and whole[1]["commits_analysed"] == 5
File History 2 commits
sha256:b4b999434d042d74edc7032f1f2cfd20fda5b3580fe85966351db5b8eef0abb3 feat(intel): every grammar on the hub, and every intel job … Opus 5.5 minor ⚠ 1 day ago
sha256:41ca619ffa23246b6e5d0cbb80733d8fec729dd65efe2acf92352370ee550972 fix(intel): bound coupling's pair counting; make the commit… Sonnet 5.5 minor ⚠ 4 days ago