fix looping with JS/TS

[Saga4]

No description provided.

[claude]

Claude finished @Saga4's task in 5m 6s —— View job

---

PR Review Summary

• Triage PR scope
• Run linting and type checks
• Resolve stale review threads
• Review code changes
• Duplicate detection
• Test coverage analysis
• Optimization PRs handled

---

Prek Checks

All checks passed. No linting or formatting issues were found. The 65 mypy errors present across the changed files are pre-existing and not introduced by this PR.

Code Review

The PR addresses two real problems with JS/TS benchmarking: stdout pipe backpressure from excessive loop iterations, and inadequate subprocess timeouts for slow JS toolchains.

JS_BENCHMARKING_MAX_LOOPS: 100k → 5k (support.py)

The rationale is well-explained in the comment: 100k iterations × ~200 bytes/marker = 20–200 MB of stdout, creating pipe backpressure that inflated wall-clock time past the subprocess timeout. 5k is a reasonable reduction that still provides stable timing data.

Timeout formula (vitest_runner.py:622, test_runner.py:1033, mocha_runner.py:541)

Old: max(120, (target_duration_ms // 1000) + 60, timeout or 120)
New: max(120, (target_duration_ms // 1000) + 120, (timeout or 60) * 5)

In production, timeout=INDIVIDUAL_TESTCASE_TIMEOUT=15 is always passed, so:

• The (timeout or 60) * 5 term evaluates to 15 * 5 = 75, which falls below the floor of 120 and has no effect
• The effective change is just + 60 → + 120 in the second term (more headroom for startup)

The (timeout or 60) * 5 sub-expression is confusing since it has no practical effect with the actual caller values. A simpler formula like max(300, (target_duration_ms // 1000) + 120) would be clearer and more directly achieve the stated goal (comparable to behavioral tests at 600s). Fix this →

get_src_code fallback (models.py:775-782)

When libcst.parse_module fails (JS/TS files), instead of returning None, the PR returns a descriptive comment stub. The intent is good — code repair gets some context instead of nothing.

However, test_src is already read at line 773 before the cst.parse_module call fails. Returning test_src directly would give the code repair API the actual test content:

try:
    test_src = test_path.read_text(encoding="utf-8")
except Exception:
    return None

try:
    module_node = cst.parse_module(test_src)
except Exception:
    return test_src  # JS/TS: libcst can't parse, return raw source

The current approach returns only metadata (function name, file name), not the actual test body, which limits the usefulness for code repair. Fix this →

Duplicate Detection

The identical total_timeout formula is repeated verbatim in all three benchmarking runners (vitest_runner.py, test_runner.py, mocha_runner.py). This is pre-existing duplication (not introduced by this PR), but extracting it to a shared helper in support.py would be a worthwhile follow-up.

No other duplicates detected.

Test Coverage

All 23 dispatch tests pass. Coverage is low (~7–13%) for the runner files due to integration dependencies (Node.js runtimes), which is expected. The test update in test_support_dispatch.py correctly reflects the max_loops constant change.

Optimization PRs

• Merged PR ⚡️ Speed up function add_runtime_comments by 112% #1762 (speed up add_runtime_comments by 112%, CI pass, mergeable)
• Merged PR ⚡️ Speed up function _generate_sqlite_write_code by 87% #1760 (speed up _generate_sqlite_write_code by 87%, CI pass, mergeable)
• Merged PR ⚡️ Speed up method JavaAssertTransformer._infer_return_type by 44% in PR #1655 (feat/add/void/func) #1757 (speed up JavaAssertTransformer._infer_return_type by 44%, CI pass, mergeable)
• Closed PR ⚡️ Speed up function _is_test_file_by_pattern by 122% in PR #1759 (merge/misc-fixes-into-omni-java) #1761 (merge conflicts)

---