`_parse_pytest_output` matches "passed" inside test names ("passed_validation")

[popywalker]

Found while testing the dynamic eval pipeline.

backend/app/code/eval/dynamic_eval.py:218-232:

passed_match = re.search(r'(\d+)\s+passed', combined)
if passed_match:
    passed = int(passed_match.group(1))

failed_match = re.search(r'(\d+)\s+failed', combined)
if failed_match:
    failed = int(failed_match.group(1))

skipped_match = re.search(r'(\d+)\s+skipped', combined)
if skipped_match:
    skipped = int(skipped_match.group(1))

Two problems:

1. Regex greediness on test names. If pytest verbose output includes a test like test_42_passed_with_warning ... somewhere in stdout, the first match for r'(\d+)\s+passed' could be the 42 in that test name, not the actual summary line 5 passed, 2 failed.

2. No anchor to summary line. Pytest's summary lines look like === 5 passed, 2 failed, 1 skipped in 3.21s ===. The right regex should anchor on the summary:

   summary_re = re.compile(
       r'={3,}\s*'
       r'(?:(\d+)\s+failed,?\s*)?'
       r'(?:(\d+)\s+passed,?\s*)?'
       r'(?:(\d+)\s+skipped,?\s*)?'
       r'(?:(\d+)\s+(?:error|warning),?\s*)?'
       r'in\s+[\d.]+s'
       r'\s*={3,}',
       re.IGNORECASE,
   )

3. (Bonus) No handling for xfailed, xpassed, errors. Real pytest output regularly contains these and they're currently treated as nothing.

Concrete demonstration: take any project with test_42_passed_smoke.py in its layout, run pytest -v against it with no actual tests passing; the parser will report passed=42 instead of 0.

Fix

Use the anchored summary regex above and prefer the last match in stdout/stderr (pytest always summarises at the very end). Replace the three independent regexes with the single anchored one.