Bug 561874 – dev-libs/liblinear is bundling libblas

• Status: RESOLVED INVALID
• Product: Gentoo Linux

[Description Justin Lecher (RETIRED) 2015-09-30 06:26:00 UTC]

Unbundle those libs

As a reminder in case you are not familiar:
https://wiki.gentoo.org/wiki/Why_not_bundle_dependencies
https://fedoraproject.org/wiki/Packaging:No_Bundled_Libraries

[Comment 1 Michał Górny 2015-09-30 07:11:51 UTC]

Would be nice if you listed at least some of the bundled libs.

[Comment 2 Michał Górny 2015-09-30 07:17:30 UTC]

Ok, I see what this is about now.

[Comment 3 Jeroen Roovers (RETIRED) 2015-09-30 08:50:19 UTC]

"Bundling" means that it is shipping and using a copy of software available elswhere. Show me that evidence.

[Comment 4 Justin Lecher (RETIRED) 2015-09-30 08:54:38 UTC]

(In reply to Jeroen Roovers from comment #3)
> "Bundling" means that it is shipping and using a copy of software available
> elswhere. Show me that evidence.

readelf -Ws /usr/lib64/librefblas.so| grep dscal_
   157: 0000000000036ef0   198 FUNC    GLOBAL DEFAULT   11 zdscal_
   161: 0000000000010530   277 FUNC    GLOBAL DEFAULT   11 dscal_

[Comment 5 Jeroen Roovers (RETIRED) 2015-09-30 09:06:40 UTC]

You mean a BLAS implementation implementing BLAS functions is evidence that another BLAS implementation should have been used?

[Comment 6 Justin Lecher (RETIRED) 2015-09-30 10:28:07 UTC]

(In reply to Jeroen Roovers from comment #5)
> You mean a BLAS implementation implementing BLAS functions is evidence that
> another BLAS implementation should have been used?

Let's put it the other way, show evidence that unbundling is the wrong thing to do.

[Comment 7 Jeroen Roovers (RETIRED) 2015-09-30 11:14:26 UTC]

No no no, the question is really simple: which BLAS implementation is being bundled here, where "bundled" is to be understood in terms of our QA policies, as in, provided by another package.

[Comment 8 Jeroen Roovers (RETIRED) 2015-09-30 11:16:29 UTC]

And to be absolutely sure you get it: bundling is about a copy of another library, not about a library implementing the same API.

[Comment 9 Michał Górny 2015-09-30 11:20:46 UTC]

Can we agree on having a flag for those of us who have to install another blas implementation anyway?

[Comment 10 Jeroen Roovers (RETIRED) 2015-09-30 11:23:37 UTC]

Sure, and I intend to introduce that USE flag, probably based on jlec's work, too, but this bug is about "bundling" and whether it should be closed as WONTFIX once we agree.

A change request for said USE flag should not be confused with allegations of QA violations.

[Comment 11 Jeroen Roovers (RETIRED) 2015-09-30 11:24:26 UTC]

Especially when there is a comrel bug that apparently hinges on the outcome of this bug report.

[Comment 12 Justin Lecher (RETIRED) 2015-09-30 11:34:53 UTC]

Comrel has nothing to do with this bug. This is about quality of packaging.

liblinear is exposing symbols specific to blas. That is QA.


Fedora and Debian (I didn't check other distros) are unbundling blas from liblinear since long.

[Comment 13 Jeroen Roovers (RETIRED) 2015-09-30 11:40:12 UTC]

You still haven't provided any evidence. And yes, that comrel bug has everything to do with it, because without your allegations of QA violations, that bug has no leg to stand on.

So prove, by code comparison preferably, which other BLAS library is being bundled here.

[Comment 14 Jeroen Roovers (RETIRED) 2015-09-30 12:00:38 UTC]

Can't fix what?

[Comment 15 Michał Górny 2015-09-30 12:01:04 UTC]

(In reply to Jeroen Roovers from comment #10)
> Sure, and I intend to introduce that USE flag, probably based on jlec's
> work, too, but this bug is about "bundling" and whether it should be closed
> as WONTFIX once we agree.

I would rather see the bug topic changed rather than RESO/INVA or the like. Just please make sure to also copy my quoting fixes which affected the existing code as well.

[Comment 16 Jeroen Roovers (RETIRED) 2015-09-30 12:33:04 UTC]

Truthfully I never want to see this bug report again if I can help it. So changing the Summary is something I couldn't live with, really.