CI: PublishTestProjectAOTModule intermittently hangs for 30 minutes on macOS #7211

What happened

On the modularpipeline (macos-latest) job, PublishTestProjectAOTModule sometimes runs until the 30-minute module timeout and is then killed:

PublishTestProjectAOTModule ✗ (30.0m) - ModuleTimeoutException
ModuleTimeoutException: PublishTestProjectAOTModule has timed out after 30m & 0s. The module did not respond to the cancellation token and was forcibly ...

The same module normally finishes in about 1.8 minutes on macOS, so this is a hang, not a slow publish. Seen on 2026-10-20 on three unrelated PRs:

• PR fix(sourcegen): pass omitted optional defaults so calls bind to the test method #7148, run 39343087633 (job 118369102723)
• PR refactor(engine): dedupe reflection generic discovery #7183, run 39374154680 (job 118468411331)
• PR test(assertions): make WaitsFor cancellation tests robust to slow CI runners #7176, run 39374557168 (job 118470432545)

Rerunning the failed job passes.

Where

tools/TUnit.Pipeline/Modules/PublishTestProjectAOTModule.cs, which runs dotnet publish with Aot=true for TUnit.TestProject (net10.0) into artifacts/aot-publish.

Expected

The publish either completes or fails fast. Worth checking:

• whether the native linker / ILCompiler step stalls under the parallel load of other modules on the macOS runner (memory pressure);
• adding ShowStandardOutput or a heartbeat so the hang point is visible in logs;
• whether dotnet publish should get --disable-build-servers or a lower ILC parallelism on macOS;
• whether the module should honour cancellation so the job fails at the timeout instead of being force-killed.

Reproduce

Intermittent on GitHub-hosted macos-latest runners during a busy queue.