Files
C.E.L_Slide_test2/tests/test_pipeline_smoke_imp85.py
kyeongmin cacc5b30db feat(#85): IMP catalog builder invariant + VP runtime gate (u1~u7)
- u1: BuilderMissingError(FitError) — narrow exception aligned with pipeline catch
- u2: load_frame_contracts catalog invariant + VP skip + CatalogInvariantError
- u3a: audit CLI I1~I3 (partial existence / declared builder / registry membership)
- u3b: audit CLI I4 (slot_payload refs vs declared/generated payload keys)
- u4: lookup_v4_candidates VP filter (lookup_v4_all_judgments raw telemetry untouched)
- u5: catalog invariant regression coverage + temp non-VP failure fixtures
- u6: mdx04 VP routing fixture tests (sw_dependency_four_problems excluded from live)
- u7: tests/conftest.py env isolation + mdx03/mdx04/mdx05 subprocess smoke

Targeted 74 PASS (12.31s). Full regression 1063 PASS (87.70s). Audit CLI clean.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-23 16:56:38 +09:00

128 lines
4.8 KiB
Python

"""IMP-#85 u7 — subprocess smoke for mdx03 / mdx04 / mdx05 pipeline runs.
These smokes exercise the IMP-#85 catalog ↔ contract ↔ builder
invariant + runtime VP gate end-to-end against real MDX inputs:
* mdx03 — non-VP rank-1 path stays clean (exit 0).
* mdx04 — the original IMP-#85 hard-crash signature
(``BuilderMissingError ... PAYLOAD_BUILDERS has no such entry``)
is GONE. u1 converted the uncaught ``ValueError`` into a
``BuilderMissingError(FitError)`` subclass; the pipeline's
existing ``except FitError`` at ``src/phase_z2_pipeline.py:4436``
catches it and the zone is routed to
``adapter_needed (skip render)``. Anything that crashes
*downstream* of that routing (e.g. layout_css zone aggregation
when all live zones are adapter_needed) is a separate axis and
out of scope for this issue (see follow_up_issue_candidates).
* mdx05 — non-VP rank-1 path stays clean (exit 0).
Each subprocess gets a unique run_id so the runs do not collide on
disk when pytest is invoked concurrently or with -x retry.
"""
from __future__ import annotations
import subprocess
import sys
import uuid
from pathlib import Path
import pytest
REPO_ROOT = Path(__file__).resolve().parents[1]
SAMPLES_DIR = REPO_ROOT / "samples" / "mdx_batch"
# Original IMP-#85 crash signature (issue body verbatim). u1 converted
# the uncaught ``ValueError`` raised from the mapper's missing-builder
# branch into a ``BuilderMissingError(FitError)`` subclass that the
# pipeline catches. The string below was the marker of the uncaught
# propagation; it must no longer appear in stdout/stderr of a mdx04
# subprocess run.
IMP85_OLD_CRASH_MARKER = "PAYLOAD_BUILDERS has no such entry"
def _run_pipeline(mdx_name: str, run_id: str, timeout: int = 240) -> subprocess.CompletedProcess:
"""Spawn ``python -m src.phase_z2_pipeline <mdx> <run_id>`` and capture I/O."""
return subprocess.run(
[
sys.executable,
"-m",
"src.phase_z2_pipeline",
str(SAMPLES_DIR / mdx_name),
run_id,
],
capture_output=True,
text=True,
timeout=timeout,
cwd=str(REPO_ROOT),
)
def _unique_run_id(prefix: str) -> str:
return f"{prefix}_imp85_smoke_{uuid.uuid4().hex[:8]}"
@pytest.mark.parametrize(
"mdx_name,prefix",
[
("03.mdx", "mdx03"),
("05.mdx", "mdx05"),
],
)
def test_non_vp_smoke_runs_clean(mdx_name: str, prefix: str) -> None:
"""mdx03 / mdx05 hit non-VP rank-1 frames; the pipeline runs to exit 0.
Non-VP rank-1 selection is the normal Phase Z path and the
primary regression guard that u1-u6 do not perturb mapper /
pipeline behaviour for non-VP routes.
"""
cp = _run_pipeline(mdx_name, _unique_run_id(prefix))
assert cp.returncode == 0, (
f"{mdx_name} pipeline returncode={cp.returncode}\n"
f"--- stderr tail ---\n{cp.stderr[-1500:]}\n"
f"--- stdout tail ---\n{cp.stdout[-1500:]}"
)
def test_mdx04_no_longer_emits_imp85_crash_signature() -> None:
"""mdx04 must no longer surface the IMP-#85 uncaught crash marker.
Before u1: missing-builder ``ValueError``
(``'PAYLOAD_BUILDERS has no such entry'``) propagated uncaught and
killed the pipeline at the mapper call site
(``src/phase_z2_pipeline.py:4411-4413``, ``except FitError``
only). After u1: the mapper raises
``BuilderMissingError(FitError)``, the pipeline catches it at the
same ``except FitError`` block, and the zone is recorded under
``adapter_needed (skip render)``.
This smoke asserts only that the original IMP-#85 marker is gone
from both stdout and stderr — downstream crashes (e.g.
``build_layout_css`` zone aggregation when all live zones are
adapter_needed) belong to a separate axis and are tracked as a
follow-up issue candidate.
"""
cp = _run_pipeline("04.mdx", _unique_run_id("mdx04"))
combined = cp.stdout + cp.stderr
assert IMP85_OLD_CRASH_MARKER not in combined, (
"IMP-#85 original crash signature still present in pipeline output:\n"
f"--- stderr tail ---\n{cp.stderr[-1500:]}\n"
f"--- stdout tail ---\n{cp.stdout[-1500:]}"
)
def test_conftest_env_isolation_active_for_ai_fallback_defaults() -> None:
"""Direct assertion that ``tests/conftest.py`` isolated the AI
fallback env vars BEFORE ``src.config`` was first imported.
With ``AI_FALLBACK_ENABLED=true`` in the live ``.env``, the
Settings default-OFF contract would otherwise be violated whenever
a developer runs ``pytest -q tests`` against a checkout that has a
live operator ``.env``. This test pins the contract to the source
of truth (``src/config.py`` defaults).
"""
from src.config import Settings
s = Settings()
assert s.ai_fallback_enabled is False
assert s.ai_fallback_auto_cache is False