Run tests with address, undefined, and thread sanitizer in CI #4
Conversation
|
Sigh, there's either a bug in libstdc++ or GCC just reports false positives (https://github.com/bcumming/uenv2/actions/runs/10110902791/job/27962015058?pr=4). This seems to be an issue all the way up to GCC 14 (https://godbolt.org/z/o6zq4TT8M) with a combination of |
msimberg
force-pushed
the
sanitizers-ci
branch
2 times, most recently
from
0a8ede8
to
cf83da7
Compare
|
I downgraded asan to lsan to avoid the false positive and at least have some sanitizers used. @bcumming is this skipped test expected to be skipped: https://github.com/eth-cscs/uenv2/actions/runs/10697698093/job/29655442215?pr=4#step:9:20? |
That test needs to be fixed - a "setup" step that creates the input data sets needs to be performed first. |
msimberg
force-pushed
the
sanitizers-ci
branch
from
cf83da7
to
c0da8cd
Compare
|
Thanks @bcumming for fixing the test setup. This should now be ready to go. |
tsan is maybe a bit unnecessary, but doesn't hurt much, and then it's there in case there will be threading in the future.
msan is not supported by GCC so leaving that out (I believe that also requires recompiling the standard library to be effective).
asan reports false positives during release mode compilation so I've kept asan in debug builds only, and added lsan for release and debug.