There is a recurrent issue with packages maintained by Tom Rix (trix) that involve BLAS/LAPACK linking. And I'm not sure what to do here, because the maintainer is clearly active but somewhat ignores my requests, but also there is an underlying problem here that has todo with the review process.

Context

We have a clear policy for BLAS/LAPACK consumers. At the beginning, linking to FlexiBLAS required some effort, because it was not so well known. Nowadays, it is recognized by cmake and it's pretty easy, although some packages still require some tuning, and this is some work I'm willing to do myself. Due to this friction (and also sometimes maintainers just forget about this stuff, it's completely normal), I have a script to regularly check that there are no random packages linking to other libraries. Now, most of the times, when this happens, I open a bug, the maintainer reacts quickly and fixes the issue (thanks, Sandro). Sometimes it requires more time, or my help via a PR. No problem.

What happened so far

It all started as described before. I detected that python-torch (in this case, a new package) was linking to BLAS or OpenBLAS, I don't recall. I filed a bug, and it took a couple of months, but it was resolved. Fine. I was a bit worried because, as a new package, it should have NOT passed the review process with this issue.

Then I was more worried because another new package, caffe, with the same maintainer, had exactly the same issue. I filed a bug. This time I got a quick response, but the fix effectively disabled BLAS, instead of linking to FlexiBLAS correctly. This is one of those packages requiring some workarounds. So I opened a PR with a proper fix. It was closed after some weeks. Then I reopened explaining that the issue wasn't fixed, but avoided. It has been ignored for 5 months now.

Today I check and there's just another package with the same maintainer, magma, and the same issue. Of course I filed the corresponding bug. But I'm a bit tired of this, as you can imagine.

The issues

There are 2 separate issues here:

• There is a package maintainer who is active but repeatedly ignores a packaging guideline.
• These 3 packages (python-torch, caffe, magma) passed the review process without complying with this guideline.

And I'm out of tools here, so I hope FESCo could provide some guidance.