Open luhenry opened 3 months ago
Note: #483 comes from a compiler bug known to be fixed in GCC12.4 and GCC13.3 (both of them upcoming releases). Should also be fixed in GCC-14.
Note: https://github.com/shibatch/sleef/issues/483 comes from a compiler bug known to be fixed in GCC12.4 and GCC13.3 (both of them upcoming releases). Should also be fixed in GCC-14.
Great! I would suggest we merge this PR with the tests disabled, and we can enable them when we know GCC 12.4 is shipping in Ubuntu repositories.
Thanks again for the PR! In principle we are never against more testing and it sounds like the way forward, but we need to be conscious of resources and maintenance cost. Is there a strong need for testing 2 versions of each compiler, if so why not more? Should we stick to the latest and maybe test older versions less frequently (weekly).
We are worried that as we increase the number of tested parameters (OS-es, compilers, LTO/no-LTO, shared/static libs, ...) the workflow will progressively become unmaintainable and possibly exceed github runners quotas. So I think it is worth planning carefully what tests we want to do and how we could break them down into smaller pipelines.
Hi! Sorry to keep you waiting here.
We might be able to integrate tests for more compilers if we re-organise and re-parametrise workflows in such a way that expensive workflows like test-cross are only triggered periodically (as opposed to on commit).
That means non-native tests will happen less often, but can be made more thorough (more os, compilers, configurations, ...).
Let's try to merge this PR first and re-organise in the coming weeks.
aarch64-gcc-12, ppc64el-gcc-12, and s390x-gcc-12 are not enabled as they are failing with https://github.com/shibatch/sleef/issues/483