[Mikola Lysenko (mikolalysenko)]

Problem

vendor::pypi_hatch::tests::hatch_on_an_absolute_path_entry_passes_the_version_gate flakes in the coverage (1) leg, which ci-ok needs. That makes it a merge-queue eviction risk.

• PR Fix hosted Go pins outranked by go.work members (#1753) #1755, run 40612541944: attempt 1 failed (job 122567365331), and the re-run passed on the same SHA.
• Panic: pypi_hatch.rs:1141:10: called Result::unwrap() on an Err value: ("pypi_hatch_unsupported", "vendored environment dependencies require Hatch >=1.2 on PATH ...")

Root cause

The test writes a fake hatch script, and the first exec of that script happens inside the production version probe. A sibling test thread that forks while fs::write holds the script open keeps a write fd until its own exec. Exec'ing the script in that window fails with ETXTBSY. require_environment_context_support_with turns any spawn error (.ok()) into the same refusal as a Hatch that is too old, so the test sees an Err. This is the same class as exec_freshly_written in the CLI tests and output_of_fresh_shim in utils::process.

Fix

fake_hatch now execs the script once through the existing utils::process::tests::output_of_fresh_shim helper, which retries ETXTBSY only. It then deletes the marker that run created. Once one exec succeeds, no process holds a write fd and none can gain one, so the probe's own spawn can't race. The only other change is making that helper pub(crate). Production code is unchanged and no assertion is weakened: planted_hatch_in_the_project_is_never_executed still checks the marker after the probe.

Proof

Stress run with --test-threads=64 on pypi_hatch::tests::hatch_on pypi_hatch::tests::planted utils::process (skipping the 10s hung/timeout probes):

build	failures
origin/main (b787454)	6 / 1500, each with the exact CI panic
this branch	0 / 3000

The full vendor::pypi_hatch and utils::process modules pass (33 tests). rustfmt --check is clean on the touched file. cargo clippy -p socket-patch-core --all-targets reports only errors that already exist on main in other files; CI's clippy job (--workspace --all-features, no test targets) is unaffected by cfg(test) code.

Removed or moved tests

None.

---

Generated by Claude Code

---

Note

Low Risk
Test-only changes (cfg(test) / test helpers); production Hatch probing is unchanged.

Overview
Fixes flaky pypi_hatch Hatch version-gate tests on Unix by avoiding a fork/exec race (ETXTBSY) when a freshly written fake hatch script is first executed inside the real version probe (spawn failures are treated like “Hatch too old”).

fake_hatch now runs the script once through the existing output_of_fresh_shim helper (ETXTBSY-only retries), clears the marker from that warmup run, and documents why the probe must not be the first exec. utils::process exposes that helper and its test module as pub(crate) so the vendor test can reuse it—no production behavior changes.

^{Reviewed by Cursor Bugbot for commit 1a61d16. Configure here.}

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

bugbot run

---

Generated by Claude Code

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 1a61d16. Configure here.}

[Mikola Lysenko (mikolalysenko)]

Labeled Ready for review at 1a61d16.

• CI: 31/31 non-skipped checks green on the head (19 skipped).
• Bugbot: reviewed 1a61d16, no findings.
• Look at: test-only change. The hatch version-gate test now runs the fake hatch script once through output_of_fresh_shim (made pub(crate) in utils/process.rs) so the probe never hits the ETXTBSY race; production code is unchanged.
• Slack announcement not sent this run (Slack send tool unavailable); a later run will retry.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Ready for review at 1a61d16ff75e001f7e737b04b78a9a918c27ef27 (burn-down agent label sweep).

• CI: 50/50 check runs green or skipped on this head; mergeable, no conflicts.
• Bugbot: reviewed 1a61d16, no unresolved findings.
• Slack announcement not sent this run (Slack send unavailable); a later run will retry.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Final review brief

What it does — Test-only fix for an ETXTBSY flake in hatch_on_an_absolute_path_entry_passes_the_version_gate. The fake hatch script is now exec'd once (retrying only ETXTBSY) right after it is written, so the production version probe never performs the first, racy exec. No production code changes.

Risk — low. #[cfg(test)] code only; the one visibility change (mod tests / output_of_fresh_shim → pub(crate)) is inside a test module.

Look here

• pypi_hatch.rs:1091-1099 — rationale comment
• pypi_hatch.rs:1113-1117 — settle exec + marker removal (keeps planted_hatch_in_the_project_is_never_executed meaningful)
• process.rs:556-571 — reused helper, retries only ExecutableFileBusy

Verified — read the full diff and surrounding tests; the marker is removed after the settle run, so the "planted hatch never runs" assertion still tests the probe. Ran cargo test -p socket-patch-core --lib pypi_hatch locally: 15/15 pass. CI 31/31 green on head; Cursor Bugbot: no issues. CHANGELOG.md untouched.

Changes I made — none.

Open questions — none.

Auto-merge is armed: approving sends it straight to the merge queue.

---

Generated by Claude Code