John Paul Adrian Glaubitz

> > it seems that the issue might be within the numpy library. > > Then type `r` (to run) and `bt` for more details about where things crashed. >...

> @glaubitz I'll see if we can configure `ninja test` to avoid similar difficulties in the future. That would be extremely useful! Thanks a lot!

> @glaubitz To better understand your workflow, would you have done a `meson build`, followed by a `meson install`, followed by the `ninja` command above? `skimage` can only be tested...

The standard way is certainly to run the `pytest` command in one way or another from the build directory.

I didn't find the documented developer workflow in the `README.md`, so I didn't know how it works. As I said, distribution maintainers and architecture porters touch a lot of upstream...

Sorry for not getting back earlier. I'm now trying to build and run on x86_64 since there might also be a regression in conjunction with GCC 14. For my dayjob...

> ``` > pip install spin > spin build > spin test > ``` While `spin build` does actually work, `spin test` does not: ``` (unstable-amd64-sbuild)glaubitz@acrux:~/scikit-image$ spin test Invoking `build`...

The `cpu_count()` problem occurs on sparc64 as well. The underlying problem is that `psutil` apparently doesn't understand the different between available CPUs and online CPUs which is not the same....

> i guess we’ll want a compile-time option that overrides the usual search logic with a fixed resources path? Yes, that would be handy. Being able to set this via...

Both tests are still failing on sparc64 for the current version of ola, see: https://buildd.debian.org/status/fetch.php?pkg=ola&arch=sparc64&ver=0.10.9.nojsmin-7.1&stamp=1757499977&raw=0 Access to SPARC machines for testing can be obtained through the GCC Compile Farm: https://gcc.gnu.org/wiki/CompileFarm...