Today I was tracking a strange report about rav1e not building/executing right on fedora's rust distribution.
I got some spurious failures in a test and only on 1.32.0, if I run the test alone it works as supposed.
Anybody got a similar problem?
@Luca Barbato are the failures consistent? Can you run in in a tool like Miri or Valgrind? The first thing that comes to mind is that you have some
unsafe code (I assume a project like rav1e has a healthy chunk of
unsafe) that is triggering UB and 1.32 just happens to change something to make it evident.
It seems consistent and an not an UB (the unsafe code in rav1e is pretty minimal btw): the test failing when run with all the others does not fail when run alone.
The spurious failing test happens on macos, the build problems on fedora seem correlated to the bindgen version used for the decode tests.
how to use Miri btw?
Miri usage is documented in https://github.com/solson/miri#running-miri-on-your-own-projects-test-suite