[marsavicc]

Motivation

Closes #11488

Windows test binaries were dying with unexplained exit codes on deep test paths. test_exec_root pointed at the build tree, and this suite's deep libc++-style layout plus a long CI workspace prefix pushed %t/%t.d past MAX_PATH. Point it at a short %TEMP%-based path on Windows instead, nothing outside lit reads from it.

Note: with the exec root now under %TEMP%, prepare_execdir()'s short-scratch-dir fallback (#10939) no longer triggers in practice because every %t.d is short enough that it returns early, so the deep tests run in their real %t.d again instead of %TEMP%\hipthreads-test\<hash>. Nothing under test/std touches files relative to the cwd, so tests are unaffected, but it isn't visible from the diff. The fallback stays load-bearing regardless: gettempdir() is environment-driven, and it's the guard that engages first if the root grows (at a 75-char exec root, vs. 85 for the executable= path)

Also added a config-time warning (test/lit.cfg) that fires if test_exec_root or the test tree ever grow deep enough to approach MAX_PATH again, computed from the actual tree rather than a hardcoded depth.

[therock-pr-bot]

✅ All Policy Checks Passed

Check	Status	Details
📝 PR Description	✅ Pass	—
⛔ Forbidden Files	✅ Pass	—
🧪 Unit Test	✅ Pass	PR does not contain code files — Unit Test auto-passed
🚫 Draft PR	🔜 To Be Enabled	—
🚩 Feature Flag	🔜 To Be Enabled	—
📊 Code Coverage	🔜 To Be Enabled	—

🎉 All policy checks passed!

📖 Need help? See the Policy FAQ for details on every check and how to fix failures.

🙋 Wish to Override Policy?

[therock-pr-bot]

🎉 All checks passed! This PR is ready for review.

[marsavicc]

These 5 runs confirm that the fix should be working fine: https://github.com/ROCm/rocm-libraries/actions/runs/35224798168/job/105264360629

Run 3 has failed because of this issue that's not on us to fix: #11489

[aleksandrasavicc]

LGTM, though I'd add a warning so a longer exec root doesn't come back silently.

[marsavicc]

@DanielMcIntosh @aleksandrasavicc, actually, since this PR makes prepare_execdir()'s TEMP-fallback branch in run.py (from #10939) basically unreachable on Windows, should we just remove it here instead of leaving dead code around? The \?-form directory creation and the executable=/lpApplicationName handling for long exe paths still earn their keep (exe paths can still exceed MAX_PATH even under a short root), but the "run in a short TEMP scratch dir" fallback and its hashing logic would have no callers left. Maybe we do it in this PR or a quick follow-up PR...

[aleksandrasavicc]

@marsavicc On removing the fallback: both are driven by the same thing, the exec root length, and the fallback triggers first: cwd 248 vs exe 260, and the exe path is always 2 chars longer. So if the root ever grows, the fallback is the half that comes back first. Dropping it while keeping executable= keeps the wrong one.
And the root isn't ours, gettempdir() reads TEMP from the environment. Either way I'd keep this out of the PR.

[marsavicc]

@marsavicc On removing the fallback: both are driven by the same thing, the exec root length, and the fallback triggers first: cwd 248 vs exe 260, and the exe path is always 2 chars longer. So if the root ever grows, the fallback is the half that comes back first. Dropping it while keeping executable= keeps the wrong one. And the root isn't ours, gettempdir() reads TEMP from the environment. Either way I'd keep this out of the PR.

Ooh you are right. Keeping both, and out of this PR.

[assistant-librarian]

TheRock Submodule Bump Activity

Newest first

• 2026-09-24 — Included in TheRock via ROCm/TheRock#8444.