Hello, Am Thu, Apr 13, 2023 at 04:37:55AM +0000 schrieb Kaelyn: > I forgot to mention that I had also tried updating to the latest numpy > (2.24.2), with the same tests failing. I agree the test failures in numpy > need to be resolved in some way for merging core-updates, since the failure > affects such a large number of common (desktop) applications on i686. My > inclination is to disable both failing tests (at least for now), and possibly > updating to the latest numpy since changing the package would trigger a > rebuild anyway. I only hesitate about disabling TestKind.test_all since I > don't know what role the Fortran bridging code plays in numpy and if the > failure is a sign of a legitimate problem instead of a platform limitation > (as the large array test failure seems to be).
I agree with your analysis: I am not overly shocked that on a 32 bit architecture creating an array of 2^31 elements fails, while on a 64 bit architecture the same test passes. Still, how about sending a bug report? Can anything be seen from the Debian package? If it is a consistent 32 bit failure, it had better be repaired upstream. For the Fortran bindings, I am hesitant as well. I think it is bad style to disable a test just because it fails ;-) Maybe someone with experience in numpy can speak up. Andreas