fix(link): #5892 — keep perry_ffi test shims out of production ext archives

[proggeramlug]

Fixes #5892.

Root cause

perry-ext-http and perry-ext-http-server ship their test_async_shims modules in production archives — the mod declaration is missing the #[cfg(test)] gate that perry-ext-net already has. The shims define the same #[no_mangle] C symbols as perry-stdlib's real perry-ffi async bridge with catastrophically different semantics:

• perry_ffi_spawn_blocking_with_reactor / perry_ffi_spawn_blocking → run the closure inline: no runtime context, no ensure_pump_registered(), so the runtime: JS-loop-driven current-thread async runtime + wait-driver with gated fast-native-drive (supersedes #5779) #5831 wait-driver never registers
• perry_ffi_spawn_async → no-op that silently drops the task

When the linker resolves these from an ext archive instead of libperry_stdlib.a (single-pass archive scan; the referencing member and the shim member sit in the same archive), the current-thread runtime is never driven: js_wait_for_event falls back to the 1 s condvar park and any in-process server+client program parks forever. The auto-opt compile path is immune because strip-dedup (#5000/#5685) localizes duplicate wrapper globals — which is why only the cargo-test/debug-driver path hangs.

Evidence (hang-lab, debug/5892-hang-lab branch)

• Hung binary: single thread, timed futex cycling every ~1 s (condvar fallback), server LISTENing, zero client connections — spawned tasks never ran (runs 28664851669, 28665408372)
• Binary-corruption and self-deadlock hypotheses eliminated (identical md5 on clean recompile; timed not infinite futex)
• With this gate: single-run repro clean + 6/6 exact cargo-test iterations clean vs hang-on-iteration-1 in three consecutive ungated runs (run 28669296187)
• A same-run A/B (ungated loop → gate → loop) is running now for final causality confirmation; will post results here

Impact

Unblocks: nightly full-workspace cargo-test (red since ~Jun 21 from this + the pre-#5831 write_end flake family), release-tag Tests gates, and restores real async dispatch semantics to any natively-compiled binary whose link picked the shims (spawn_async dropping tasks silently is not test-only behavior anyone wants in the field).

Summary by CodeRabbit

• Chores
  • Updated test-only code to compile conditionally in two components.
  • No user-facing behavior, APIs, or runtime functionality changed.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 7a952d65-79d6-455c-a056-b11f6f851821

📥 Commits

Reviewing files that changed from the base of the PR and between edd608a and 7a21506.

📒 Files selected for processing (2)

• crates/perry-ext-http-server/src/lib.rs
• crates/perry-ext-http/src/lib.rs

---

📝 Walkthrough

Walkthrough

This PR adds #[cfg(test)] attributes before test module declarations in two files: perry-ext-http-server/src/lib.rs and perry-ext-http/src/lib.rs, restricting compilation of test code to test builds only.

Changes

Test compilation gating

Layer / File(s)	Summary
Add cfg(test) attributes crates/perry-ext-http-server/src/lib.rs, crates/perry-ext-http/src/lib.rs	Inserted #[cfg(test)] attributes before test module declarations so test-only code compiles only under test configuration.

Estimated code review effort: 1 (Trivial) | ~2 minutes

Related issues: None of the changes address the linked issue (#5892) regarding cargo-test hangs in issue_4903_listen_callback_deferred; these are unrelated single-line test-gating edits.

Suggested labels: trivial, chore

Suggested reviewers: None required given the trivial scope of this change.

_{🐰 A whisker-thin change, so small and neat,
Two lines of #[cfg(test)], a tidy feat,
No logic stirred, no bugs to chase,
Just gates for tests kept in their place.}

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Description check	⚠️ Warning	The description is detailed, but it does not follow the repository template sections for Summary, Changes, Related issue, Test plan, or Checklist.	Rewrite the PR body to match the template, adding Summary, Changes, Related issue, Test plan, and Checklist sections.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly states the core fix: gating perry_ffi test shims out of production archives.
Linked Issues check	✅ Passed	The change matches #5892 by adding #[cfg(test)] to the test shim modules in both HTTP crates, which is the requested fix.
Out of Scope Changes check	✅ Passed	The diff is narrowly scoped to test-only module gating in two files and introduces no unrelated code changes.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/5892-test-shims-cfg-gate

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[proggeramlug]

Final lab results (runs 28669296187, 28671098729): with archives built by the lab's combined cargo build invocation, the ungated link happened to pick the real stdlib shims (bug-state loop 0/4 hangs) — the shadowing is sensitive to how the wrapper/ext archives are built (test.yml's per-package serial loop produced the hanging member order in rounds 2–4 and in every real nightly/preflight). The gate ends the sensitivity outright: after it, perry_ffi_spawn_blocking_with_reactor/spawn_async have exactly one definition in the workspace (ext-net's copy was already #[cfg(test)]-gated), so no link order can resurrect the shadow. Definitive validation = full test.yml on the merged tip, which is the next step.