The failures only happens on aarch64.
There has now been two builds in koschei since the introduction of gcc-c++-16.0.1-0.7.fc45.aarch64. The set of failing tests are not exactly the same in the two builds, but overlap.
The failing tests compare two values where the high bits in one of the values are expected to be 0, but instead seem to contain random data. E.g,
3014590463 (0xB3AEFFFF) is compared to 65535 (0x0000FFFF)
3014524933 (0xB3AE0005) is compared to 5 (0x00000005)
Or compare two values where the sign bit is flipped. E.g.
-32767 (1000 0000 0000 0001) is compared to 1 (0000 0000 0000 0001)
-32726 (1000 0000 0010 1010) is compared to 42 (0000 0000 0010 1010)
32726 (0111 1111 1101 0110) is compared to -42 (1111 1111 1101 0110)
See the logs in koschei for more examples.
These test failures where not seen with gcc-c++-16.0.1-0.5.fc44.aarch64.
Reproducible: Always
Steps to Reproduce:
1. Build root for aarch64 in rawhide
Actual Results:
New test failures not seen with earlier versions.
Expected Results:
No test failures.