[rmcdaniel]

Part of #325 (CI diagnosability).

The artifact summary keeps the last 64 KiB of each test report. The comparator report is about 450 KiB, so #826's one failing test was missing from the retained summary. It was reported early, before the tail. Finding it took a local reproduction of the exact test list.

Change. When a test summary (*-tests.txt) is cut to its tail, artifact-summary.mjs also writes <report>.failures.txt.

• It contains every TAP not ok line with its indented diagnostic block, up to the block's ....
• It stays within the same per-file and total limits, and is recorded with derived_from and the failure count.
• A whole report or a passing report gets no excerpt.

Checked: ServerParityArtifactSummaryTest passes 8/8. The new case puts a failure ahead of a 4,000-test passing tail, and also checks a passing report and a small failing report.

🤖 Generated with Claude Code

[rmcdaniel]

Resolved at b2b6e75: independent follow-up verification. The original 53a61f5 findings below are preserved.

Review at 53a61f5ade27cbd57cddae3acb225d53a81bcb1f found two regressions to address before merge. The existing artifact-summary suite independently passes 8/8.

1. The new failure scan loses the bounded-input behavior. Line 97 rereads the entire report with readFileSync(..., 'utf8'), then splits the whole string and accumulates all failure blocks before truncating the output. The 64 KiB/file and 2 MiB/content limits therefore do not bound the scanner's memory use. I compared a zero-allocation sparse 3 GiB report: the baseline retains only 65,536 bytes; the new path attempts the whole-file read. A controlled read guard intercepted that request, so this check did not allocate/read 3 GiB or establish an actual OOM. Please scan with bounded chunks/line/block buffers and bounded excerpt accumulation, with a regression proving large reports preserve those bounds.

2. Derived outputs make omitted-file accounting negative. With a single comparator-tests.txt containing one early TAP failure followed by more than 64 KiB of passing lines, the result is scanned_files=1, two retained entries (tail plus derived excerpt), and omitted_files=-1 at line 114. No source file was omitted. Count retained source files separately from derived entries and cover this one-source case.

Both checks used isolated offline Node 22.23.3. No owner-branch changes or Rust execution.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[rmcdaniel]

Independent follow-up at b2b6e7531919c6512163d6a06ba00ff8318c5e63 resolves both findings in 6461500906.

• Artifact-summary suite: 9/9 passed, including the 256 MiB sparse-report regression.
• Repeated the whole-file-read guard against a 32 MiB sparse input: collection succeeds, retains 64 KiB, and never requests a whole-file read. Reviewed reads in 64 KiB chunks, bounded pending lines/excerpt accumulation, UTF-8 decoding and O_NOFOLLOW.
• Original one-source accounting case now reports scanned_files=1, two retained entries and omitted_files=0.
• Additional controls preserve a UTF-8 character split across the 64 KiB chunk boundary and count all 7,000 failures after the excerpt fills, retaining 65,533 bytes. That separate 32 MiB control process peaked at 55,964 KiB RSS; this is a scoped measurement, not an arbitrary-input memory benchmark.

All checks ran with cached Node 22.23.3 in isolated offline containers. No Rust execution or owner-branch changes. No further finding in this correction; normal hosted checks still apply.

Hosted PR qualification independently verified at 2026-10-28 06:10 UTC, same b2b6e753: all 57 head checks are terminal, 56 successful / 1 skipped. The skipped check is Candidate cancellation and prepared-local source qualification. Shared Server fixtures40639629907, PHP40639629800, Public Boundary40639629810 and Sonar40639629835 all passed that exact head. The PR is OPEN/ready with a clean mergeable state. Normal merge-group qualification remains required.

Combined qualification and source delivery are complete: queue/merge commit dc46dc842988b54fab2c8fe7b513d8e39a24a244 passed Shared40643095197, PHP40643095237 and Boundary40643095158. GitHub confirms #833 merged at 2026-10-28 06:40:27 UTC. Both original review findings are corrected in the merged source.