Open 0xTim opened 4 years ago
Even though the exit code is different, this is probably related to SR-12424. Can confirm the tests work fine from the command line in 5.3.
Here's the binary as requested swift-cryptoPackageTests.xctest.zip
Hmm, taking a look at the crash report and the binary I don't immediately think this is the same as SR-12424. It manifests in the same situation, and may have similar causes, but the manifested behaviour is different. In particular, seeing a SIGILL from this code (entirely assertion free, with no ud2
to be found) is a bit alarming. I think I'll need to try to get a reproducer.
Ok, this is bizarre: I can reproduce a problem, not exactly this one. I am getting a SIGSEGV somewhere entirely else in the code, but in the same test. ASAN is clean (and indeed prevents the crash from happening at all, which suggests a stack smasher).
Note that I can't reproduce this when using Swift 5.3, just as @0xTim cannot, so that's a decent workaround for now while I dig in.
As a side note for future self: an update to BoringSSL does not fix this issue, so that's not going to be a suitable workaround. I didn't really think it would, but worth noting explicitly.
New Issue Checklist
Expected behavior
Test to execute and pass successfully
Actual behavior
Tests exit with signal code 4. Output is:
Steps to reproduce
CRYPTO_IN_SWIFTPM_FORCE_BUILD_API
flagswift test
from the command lineTests are fine when run from Xcode
Swift Crypto version/commit hash
f8b6e4e17e06eb9f3ede1b84466430137424d8e8
Environment
Stand in the root of the repo and run
./scripts/environments.sh
and paste the output belowCrash report: