Bug#813128: Bug#814943: mpi-default-dev: provide the list of architectures for each MPI implementation (Was: Bug#813128: transition: openmpi)
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA256
Follow-up to 814943@bugs.debian.org
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=814943
- -------- Message transféré --------
Sujet : Bug#814943: mpi-default-dev: provide the list of architectures
for each MPI implementation
Date de renvoi : Tue, 16 Feb 2016 21:21:01 +0000
De (renvoi) : Thibaut Paumard <thibaut@debian.org>
Pour (renvoi) : debian-bugs-dist@lists.debian.org
Copie (renvoi) : mattia@debian.org, Debian Science Team
<debian-science-maintainers@lists.alioth.debian.org>
Date : Tue, 16 Feb 2016 22:19:06 +0100
De : Thibaut Paumard <thibaut@debian.org>
Répondre à : Thibaut Paumard <thibaut@debian.org>, 814943@bugs.debian.or
g
Organisation : Debian GNU/Linux
Pour : Debian Bug Debian BTS submit <submit@bugs.debian.org>
Package: mpi-default-dev
Version: 1.2
Severity: wishlist
It has been discussed in
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=813128
that some MPI packages build in two flavours (openmpi and mpich) and
need to know at upload time for which architecture each implementation
is available.
It has been proposed to add two variables to
/usr/share/mpi-default-dev/debian_defaults for this purpose.
Last iteration of this discussion below.
Mattia, it looks like there is a misunderstanding: in your commit,
OPENMPI_ARCHITECTURES and MPICH_ARCHITECTURES only list the
architectures for which each implementation is the default. The
feature that I would need for e.g. the yorick package is the list of
architectures on which each implementation is available. This is what
I currently check and hardcode by hand. I guess this is also what
other people do when they provide packages with distinct names for
each flavour.
Kind regards, Thibaut.
Le 16/02/2016 19:37, Mattia Rizzolo a écrit :
> On Tue, Feb 16, 2016 at 11:37:19AM +0100, Thibaut Paumard wrote:
>>> then mpi-defaults would need a sourceful uploads every single
>>> time a new architecture is added (and we want to support MPI
>>> there and openmpi builds), and also suddenly file a dozen RC
>>> bugs (as all packages using such a system would start to fail).
>>> Yes, we can do it, though.
>
> See
> https://anonscm.debian.org/cgit/debian-science/packages/mpi-defaults.g
it/commit/?id=07ef8a6
>
>
https://anonscm.debian.org/cgit/debian-science/packages/mpi-defaults.git
/commit/?id=4fa28c2
> https://anonscm.debian.org/cgit/debian-science/packages/mpi-defaults.g
it/commit/?id=d9656b2
>
>
>> So, what we want if to render RC buggy some packages that need a
>> source upload whenever OPENMPI_ARCHITECTURES or
>> MPICH_ARCHITECTURES change.
>
> Somebody needs to do that. I can also have mpi-defaults provide a
> script to be called by the packages at build time, if somebody
> provides it.
>
>> An easier way would be for those packages to have a versioned
>> dependency on mpi-default-dev and bump this version when either
>> variable changes, e.g.
>>
>> Build-Depends: mpi-default-dev (>= 1.3), mpi-default-dev (<<
>> 1.4~)
>
> umh, looks messy.
>
>> This is assuming the minor part of the version of
>> mpi-default-dev changes when either variable changes. The version
>> can then have also a micro digit, to allow for new versions that
>> don't change these variables
>
> in the past mpi-defaults was binNMUed to change defaults; don't
> rely on that.
>
>> Actually a versioned dependency seems required anyway since you
>> know your new package will FTBFS with earlier versions of
>> mpi-default-dev.
>
> *shrugs*
>
>> The only thing is that you can also predict that later versions
>> of mpi-default-dev will break your package.
>>
>> This discussion getting off-topic for this bug, should we move
>> somewhere else?
>
> indeed. probably better suited for a mpi-defaults bug; feel free to
> open one and report a summary of what said here.
>
- --
debian-science-maintainers mailing list
debian-science-maintainers@lists.alioth.debian.org
http://lists.alioth.debian.org/cgi-bin/mailman/listinfo/debian-science-m
aintainers
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2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=z+Ab
-----END PGP SIGNATURE-----
Reply to: