Update CI infrastructure #17
Conversation
|
This is mostly formulaic, but the UB sanitizer turned up some issues in 64-bit tests. |
|
There is also an issue with tests that don't have any "real" google tests, but only rapidcheck tests. They don't play well with gtest_discover_tests (I assume reporting no tests) - the result is that they will rebuild every build... adding at least one (possibly degenerate) "plain test" would fix this. |
Can you share which tests had this issue? Was it an issue in the test itself using 64-but primitives, or in safe arithmetic? If a test hit 64-bit overflow for negation, that should fail the test because the big_integer should have the correct value while the 64-bit primitive will overflow or have other UB. Right? |
|
It is a problem with the tests, not the library. And the test doesn't fail; it triggers UBSan. For example https://github.com/intel/safe-arithmetic/blob/main/test/safe/big_integer/detail/plus.cpp#L37 when the generated values are such that |
|
Looks like I need to update the branch protection rules on the repo. 😅 |
|
Should be building now... clang-format reordering headers produced a failure I guess. |
|
Tests are "passing" (see ASan/TSan tests), but failing because of CTest interaction with rapidcheck & the compile-fail tests. And of course the UBSan tests have a couple of issues. I'm assuming this isn't holding up the actual merge though and we can fix this all up in future PRs? |
Some tests cause integer overflow: they need to be guarded by `RC_PRE`. This fixes signed 64-bit negation, addition and subtraction. Multiplication still has this problem.
Head branch was pushed to by a user without write access
clang-tidy is temporarily turned off, to be turned on in a separate fix.
2 participants
No description provided.