Issue 4577 - Add GitHub actions #4625
Conversation
|
It doesn't show all checks because I split the workflows. And for some (security?) reasons GitHub doesn't execute additional workflows that are not yet merged. Only modified workflows. I will post a link with a full run from my fork. |
|
Hmm, switching to F32 didn't help, test artifacts are still not useful. I will set this to WIP until I fix ASAN reports. |
vashirov
added
the
work in progress
|
Ok, now there are memory leaks reports. But I have another problem now and need your opinion. I propose to run with |
|
Forgot the link: https://github.com/vashirov/389-ds-base/actions/runs/574633583 |
vashirov
removed
the
work in progress
|
I think we can have fast_unwind_on_malloc=1, because knowing that any test has failed indicates to us to investigate further or to look at the nightly which does have the ability to generate the exact report. So I'm okay with this :) |
Description: * Add ASAN variant for tests * Add HTML report for pytest, that contains ASAN/LSAN reports Relates: #4577 Reviewed by: @Firstyear, @droideck (Thanks!)
| @@ -13,7 +14,15 @@ | |||
| suites += [repl_test.replace('dirsrvtests/tests/suites/', '') for repl_test in repl_tests] | |||
| suites.sort() | |||
|
|
|||
| suites_list = [{ "suite": suite} for suite in suites] | |||
| matrix = {"include": suites_list} | |||
| variants = ['gcc', 'gcc-asan'] | |||
There was a problem hiding this comment.
gcc-asan doesn't work with rust, so we can't use this. We need clang-asan to be consistent, especially so to match the correct llvm versions.
There was a problem hiding this comment.
Ok, I will replace gcc-asan with clang-asan.
There was a problem hiding this comment.
@Firstyear, clang with -fsanitize=address fails to compile: https://github.com/vashirov/389-ds-base/runs/3554323865#step:6:3541
Any ideas?
There was a problem hiding this comment.
I found that I filed an issue for this some time ago #4596
I feel that clang-asan would require more efforts than I expected, so I'd like to to keep gcc-asan for now and add clang-asan in a follow up PR.
There was a problem hiding this comment.
Didn't we change how lto was managed here to make it work? I'm building and using clang-asan just fine?
The issue with gcc-asan is that I've noticed more and more that gcc asan and rust's llvm backed asan aren't compatible and can not be mixed, which has led to failures to build. That's why I've been using clang now so that they both have the same llvm backend and asan versions.
There was a problem hiding this comment.
No, we didn't :( I couldn't make LTO builds with clang and then I switched to other things, forgetting about the issue.
I need some help from you regarding Rust + LTO, I will put the details in the related issue.
I haven't encountered any failures myself. Ideally I'd like to have the full matrix (gcc, gcc-asan, clang, clang-asan), so that we can catch these build issues too. But I'm hitting GH actions limit of 256 jobs in the matrix. I have some ideas how to circumvent this with composable actions, will try it next week.
There was a problem hiding this comment.
Okay, I'll look at the LTO issue later, but I think we can't do gcc-asan + rust at all. I've had so many conflicts between the two ASAN apis as they exist, and a bunch of false positives at this point, that I don't think it should be considered reliable.
Description:
Relates: #4577
Reviewed by: