Run the testsuite with ThreadSanitizer on a PR when label run-thread-sanitizer is added#12904
Conversation
|
Could you add a Changes entry? |
|
Done. Thank you for the review! |
|
I like the overall change (i.e. the ability to turn on tsan testing on a PR), but it's created a slightly unfortunate effect that it's no longer immediately clear that all checks passed, because the list of checks doesn't collapse down when there are skipped items. It caused me a "mild panic" on another PR because I'm used to seeing the "skipped check" mark only because another check has failed (for example, if the debug-s4096 job fails, the normal job will cancel - so the only immediately visible item is the "cancel"/"skip" mark, and you have to scroll down to find the red cross next to the debug-s4096 job) However, I have some thoughts on that, and will propose that in a separate PR 🙂 |
|
Sorry about that! I have little experience with Github Actions, so I didn’t anticipate this issue. |
Run the testsuite with ThreadSanitizer on a PR when label run-thread-sanitizer is added (cherry picked from commit 4a70c02)
(cherry picked from commit 4a70c02)
(cherry picked from commit 4a70c02)
(cherry picked from commit 4a70c02)
(cherry picked from commit 4a70c02)
(cherry picked from commit 4a70c02)
(cherry picked from commit 4a70c02)
(cherry picked from commit 4a70c02)
(cherry picked from commit 4a70c02)
(cherry picked from commit 4a70c02)
(cherry picked from commit 4a70c02)
(cherry picked from commit 4a70c02)
(cherry picked from commit 4a70c02)
(cherry picked from commit 4a70c02)
This PR proposes the possibility to run the full testsuite with ThreadSanitizer on any PR by tagging it with
run-thread-sanitizer.This is a good compromise between running TSan on all PRs and never running it. This way, the label can be added only to PRs that could use the added confidence. The TSan CI takes roughly 45 minutes.
Ideally, this job would run the full testsuite with both the normal runtime and the debug runtime, but for now I have commented out the latter, because of the races reported in #12902 which would generate a lot of noise.