Re: numba

[Diane Trout <diane@ghic.org>]

On Sun, 2024-07-21 at 09:37 +0200, Andreas Tille wrote:
> Hi Diane,
> 
> Am Sun, Jul 07, 2024 at 03:50:50PM -0700 schrieb Diane Trout:
> > 
> > I was wondering how to do this...
> > 
> > On Salsa the i386 test runner failed with this test case, which
> > seems
> > very much like a 32-bit issue.
> 
> Since you did not got any answer on this.  If this problem occures on
> i386 *only* we might start a discussion whether we want to keep i386
> support for numba and its rdepends.  There might be some applications
> for arm32 bit in scientific devices but as far as I understood Debian
> is

It does look like it's a 32bit issue. I see the same erroron armel too.

https://ci.debian.net/packages/n/numba/unstable/armel/49527832/#L3098

>  
> > I'm just uploaded numba_0.59.1+dfsg-2 with Build-Depends:
> > architecture-
> > is-64-bit. Hopefully that gets further through the builder.
> 
> Ahhh, well, this seems to be a sensible decision even excluding 32bit
> arm which is fine for me as well.  It might mean a chain of removal
> requests of several rdepends but we need to do this sooner or later
> anyway.

Maybe I can look to see if I can figure out a solution, but I kind of
want to see if I can get numba working on x86 and arm64 first.

And right now, there's test failures on the CI server that I can't
reproduce locally.

The upstream test harnesses mention issues with many core machines, and
capped the number of simulatenous test processes to 16 on ppc64.

I pushed a -3 version that tries that limit on x86_64, but it hasn't it
made to CI yet.

Diane