[jefferyb]

Summary

Prevent intermittent POSIX SIGPIPE (exit 141) failures in post-code when extracting the first commit's subject line on multi-commit branches under set -eo pipefail.

Problem

In scripts/post-code.src.sh, commit subject extraction was performed using:

first_commit_subject=$(git log "${base_sha}..${head_sha}" --pretty=format:%s | head -1)

Under set -eo pipefail, when head -1 reads the first line and terminates immediately, the pipe is closed. If git log buffers output or writes additional data after the pipe has closed, the OS raises SIGPIPE (signal 13), causing git log to exit with status 141. Under pipefail, this fails the entire pipeline and terminates post-code.sh unexpectedly.

Solution

1. Replace head -1 with sed -n '1p', which consumes the first line and allows the stream to drain cleanly without prematurely closing the pipe descriptor or triggering SIGPIPE.
2. Add a regression test in scripts/post-code-test.sh simulating multi-commit output under set -eo pipefail to ensure zero exit code even with pipe buffer saturation.
3. Rebuild bundled scripts/post-code.sh via make script-build and verify bundle parity (make check-bundle).

Verification

• Ran scripts/post-code-test.sh — verified git integration and SIGPIPE pipefail regression tests pass cleanly.
• Ran make check-bundle — verified 0 diff and clean bundle parity.
• Tested under bash 3.2 (macOS) and bash 5.x.

[qodo-code-review]

PR Summary by Qodo

Prevent SIGPIPE during multi-commit PR title selection

🐞 Bug fix 🧪 Tests 🕐 10-20 Minutes

AI Description

• Drain commit history while selecting the first subject, preventing SIGPIPE failures under
 pipefail.
• Add a large-history regression test that verifies successful extraction and preserves
 single-commit coverage.
• Rebuild the bundled script to match the source change.

Diagram

graph TD
  Count{"Multiple commits?"} -->|Yes| History["Reverse git log"] --> First["Drain with sed"] --> Subject["Commit subject"] --> Title["PR title"]
  Count -->|No| Latest["Latest commit"] --> Subject

Loading

High-Level Assessment

Keep the stream-draining sed pipeline: it preserves the existing oldest-commit selection with a minimal change. Resolving the first commit separately would add commands and complexity without a clear benefit.

Files changed (3) +28 / -6

Bug fix (2) +2 / -2

post-code.shMirror the SIGPIPE fix in the bundled script +1/-1

Mirror the SIGPIPE fix in the bundled script

• Replaces 'head -1' with 'sed -n '1p'' in the bundled multi-commit subject pipeline, matching the source script.

scripts/post-code.sh

post-code.src.shDrain multi-commit subject output before selecting the first line +1/-1

Drain multi-commit subject output before selecting the first line

• Uses 'sed -n '1p'' instead of 'head -1' so Git can finish writing its output without a prematurely closed pipe under 'pipefail'. The single-commit path is unchanged.

scripts/post-code.src.sh

Tests (1) +26 / -4

post-code-test.shCover first-subject extraction under pipefail +26/-4

Cover first-subject extraction under pipefail

• Updates the real-Git integration test to use 'sed', then adds 120 padding commits and checks that extraction returns the first subject with exit code zero under 'set -eo pipefail'. Saves the initial feature commit so the single-commit case can still be tested after the larger history is created.

scripts/post-code-test.sh

[qodo-code-review]

Code Review by Qodo

🐞 Bugs (0) 📘 Rule violations (0) 🔗 Cross-repo conflicts (0) 📜 Skill insights (0)

1. New pipefail regression test cannot catch a revert of the fix ✓ Resolved

Resolving commit: d2bc837

Resolution reason

Adds a static grep check over both post-code.src.sh and post-code.sh that fails if COMMIT_SUBJECT
uses head - instead of sed -n '1p', and changes the pipefail test to eval the production
COMMIT_SUBJECT line from POST_SCRIPT rather than an inline re-implementation.

Description

The new 'multi-commit-git-integration-pipefail-safe' block in scripts/post-code-test.sh runs its
own inline git log ... | sed -n '1p' pipeline and never invokes post-code.src.sh or the bundled
post-code.sh. If someone reverts the production line back to head -1, this test still passes
because it only checks the inline sed pipeline.

---

2. Test suite gains 120 commits on every run ✓ Resolved

Resolving commit: d2bc837

Resolution reason

Reduced the padding loop from 120 to 12 git commits by using 8KB subjects per commit, keeping total
subject output above 64KB so the SIGPIPE regression check remains meaningful, cutting per-run commit
overhead.

Description

The new loop creates 120 extra commits, each with a roughly 600-character subject, on every run of
post-code-test.sh. Each git commit spawns a process and writes objects, which adds measurable
runtime to the CI test job.

---

Context sources

✅ Compliance rules (platform): 55 rules

✅ Skills: 4 invoked

  code-review

  code-implementation

  pr-review

  docs-review

✅ Cross-repo context — repo relationships

Review mode: Auto: 🚀 Fast: Localized shell pipeline fix with focused regression coverage and limited blast radius.

Tip of the day

💡 Did you know, you can start a comment with 'qodo' or '@qodo' to chat about any finding

More tips ↗ | Customize Qodo ↗ | Qodo docs ↗

[Benkapner]

suggestion - should we use eval "${_mc_pipefail_line}" || exit $? here? if the pipeline prints the right subject and then exits 141, the printf below can still make this check pass.

[jefferyb]

Great catch — thanks @Benkapner! Because the outer || _mc_pipefail_exit=$? suppresses set -e even inside set -eo pipefail command substitutions, a failing eval previously wouldn't exit before printf.

Fixed in 1df3347ad by adding || exit $? after eval "${_mc_pipefail_line}", and verified with shellcheck --severity=warning, make check-bundle, and scripts/post-code-test.sh in both source and bundled modes.

[waynesun09]

MEDIUM: Reverting the fix aborts the whole test suite instead of reporting FAIL

This file runs under set -euo pipefail (line 10). If the sed -n '1p' pipeline is gone from COMMIT_SUBJECT=, for example after a revert to head -1, this grep -m1 exits 1 and errexit kills the script on this line. The two multi-commit-script-uses-sed FAIL lines print first, but multi-commit-git-integration-pipefail-safe never reports. Every later assertion in the file (single-commit, PR body, gitleaks, signoff and more) is skipped, cleanup does not run, and no summary is printed.

The behaviour test also finds its pipeline by grepping for the sed text, so it never runs a head -1 pipeline. Nothing shows that the >64KB fixture actually reproduces exit 141.

This is separate from the earlier thread on this test, which is fixed. That thread said the test did not run the production line at all.

Suggestion: guard the extraction:

_mc_pipefail_line="$(grep -m1 -E "COMMIT_SUBJECT=.*sed -n '1p'" "${POST_SCRIPT}" || true)"

If the result is empty, print FAIL: multi-commit-git-integration-pipefail-safe (COMMIT_SUBJECT pipeline not found) and increment FAILURES instead of running eval. Optionally, add a control run of the same git log ... | head -1 pipeline and expect exit 141. That shows the fixture really fills the pipe buffer.

[jefferyb]

Fixed in 8662279 (8662279).

• Guarded the grep -m1 -E "COMMIT_SUBJECT=.*sed -n '1p'" "${POST_SCRIPT}" || true lookup so a reverted/missing pipeline fails multi-commit-git-integration-pipefail-safe cleanly instead of aborting the test script via errexit.
• Extracted _eval_commit_subject_pipefail and added a control assertion (multi-commit-git-integration-control-head-sigpipe) replacing sed -n '1p' with head -1 and verifying exit code 141 (SIGPIPE), confirming the >64KB commit fixture reproduces the pipe overflow failure.
• Verified with shellcheck --severity=warning scripts/post-code-test.sh, make check-bundle, SCRIPT_TEST_TARGET=source bash scripts/post-code-test.sh, and SCRIPT_TEST_TARGET=bundled bash scripts/post-code-test.sh (all passed).