x86_64 blocks i386?

[Michael Schwendt]

On Mon, 21 Mar 2005 14:28:40 +0100, Nils Philippsen wrote:

> > No AMD64 test machine running F4T1+Extras, debugging x86_64 build or
> > run-time problems can become an impossible mission.
> > 
> > No AMD64 enthusiasts who would be looking into announced blocker bugs in a
> > timely manner, the requirement for packages to build on more than the h/w
> > architecture the packager volunteered to support, is dangerous.
> > 
> > During FC2 and FC3 test periods, the fedora.us requirement for packages to
> > build not just for the last stable release but also compile fine with
> > current FC Development added a significant hurdle already. It is wishful
> > thinking, that community contributors not just follow FC Devel or early
> > test releases, but also can test and debug on more than one h/w arch.
> 
> Hmm. I agree that we can't expect everyone to be able to debug on every
> arch, be it due to missing hardware, experience or both. Still I would
> be in favour of a rule that before anything hits stable trees it must be
> built on all (primary -- for whatever that is, we haven't as many arches
> as Debian yet) architectures. We don't want a situation where Extras on
> i386 is cool but sucks on another platform.

And with that we're back at the initial problem, except that it has
become completely unrelated to gpgme[03]'s failure in mach. ;)

It looks as if a rule is needed. For new packages, build failure on one
arch blocks builds for all archs? No ExcludeArch permitted for new
packages anymore? That alone would be quite rigorous and lead to
exclusion of packages which are just not designed to work on non-i386
(e.g. due to endianess related implementation issues or weird use of
C-style casts). What happens if no resources are available to look into
such a problem (and no upstream developers either)? The same can happen
during big[ger] version upgrades.

As it is currently, the majority of packages is prepared and tested on
i386 and rebuilt for x86_64 (and likely ppc[64] some day besides dwmw2's
private builds). If no arch-specific co-maintainers exist, the packages
built for x86_64 are fully untested until some users tries them out. And
it has happened before, that such an untested built either crashes right
at the start (e.g. geomview) or at a later point (e.g. anjuta).

Trying to support multiple archs equally from the start is an admirable
goal, but just not feasible until arch-specific co-maintainers kick in
(or the built stuff "just works" and no difficult issues come up).