The ranges.transform.pass.cpp often times out on CI for AIX (32-bit and 64-bit)
only. Mark the test as UNSUPPORTED for AIX for now. It should be looked into in
the future.
Details
- Reviewers
ldionne var-const Mordante philnik - Group Reviewers
Restricted Project - Commits
- rG3583826bb52a: [libc++][test] Mark ranges.transform.pass.cpp UNSUPPORTED for AIX
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
The test is just long. So either the test should be made shorted somehow or the timeout for AIX should be longer. XFAILing it doesn't solve the problem in any way. It's just the 50/50 chance of passing reversed to a 50/50 chance of failing.
I see. So, we could bump the timeout set by run-buildbot to bump it up from the current 1200 seconds, but that's awfully high already. Have you already looked into why it's taking so long (and only on AIX?)? Is it spending so much time compiling (can look with -ftime-trace and ClangBuildAnalyzer), or something else?
The test always takes a long time to run. On my machine it's 47 seconds. I think the average test takes more like one to two seconds. The problem is that we instantiate the actual test with 7 * 7 * 7 ( = 343) iterator combinations. I don't know why it takes twenty minutes on AIX though. I guess something isn't well optimized in the compiler and that gets multiplied 343 times. That's probably something someone from AIX should look into. @DanielMcIntosh-IBM @SeanP @daltenty
I would prefer to have the test fixed for AIX. However due to the recent number of false positive CI failures I don't mind to disable it temporary on one platform.
libcxx/test/std/algorithms/alg.modifying.operations/alg.transform/ranges.transform.pass.cpp | ||
---|---|---|
11 | This won't work, now the CI will fail when the test finishes within 1200s. Instead use UNSUPPORTED. |
libcxx/test/std/algorithms/alg.modifying.operations/alg.transform/ranges.transform.pass.cpp | ||
---|---|---|
11 | This fails to build due to the explanation comment being part of the code parsed by lit. |
LGTM, but agree with @philnik that the underlying issue should be fixed.
So I leave the final approval to him.
I don't think we want to do this -- I increased the timeout on our CI bots to try to avoid this issue. I think we should revert this now that I've bumped the timeout duration (hopefully enough to get rid of these spurious failures).
This won't work, now the CI will fail when the test finishes within 1200s. Instead use UNSUPPORTED.
Please add a comment why the test is disabled.