Re: Debian doesn't have to be slower than time.
Colin Watson wrote:
> One thing I've noticed is that packages sometimes do build in a standard
> environment, just subtly differently. This is both very hard to detect
> and very bad. Take one of my own packages, groff; it was OK when built
> on my machine (and the previous maintainer's machine, apparently) with
> lpr installed, but the -l option didn't work if lpr hadn't been
> installed on the build machine. How do you propose to detect this?
If someone can find an answer aside from letting enough people hammer on
it to test it, my hat's off to them.
> That encourages people not to test the packages they build, which is
> even worse. (In fact, it makes it *impossible* for them to test the
> exact .deb they've built prior to upload!)
This can go either way. Right now we have the potential for well-tested
packages in one architecture that the maintainer uploads (typically
i386) and the near certainty that the other architectures will still get
untested packages. So i386 is probably slightly getting better testing of
packages, but then, it's getting better testing anyway since so many of
our users use that architecture.
If instead everything was autobuilt (in very similar ways across
architectures), i386 would indeed lose a little bit of testing and
assurance of correct packages, but on the other hand when some bug like
the groff -l bug happened, it would affect all architectures equally,
including i386, and so it would be more likely to be quickly noticed by
one of our many i386 users.
> A better idea might be to have a farm of build machines to which people
> could submit packages, get back the .deb, and test and upload that. I'm
> not fully convinced this is viable (encouraging people to build
> everything in sbuild and pbuilder is probably a better solution), but
> it's better than forbidding binary uploads.
Perhaps rather than forbidding binary uploads we should think about at
least allowing source-only uploads for maintainers who would prefer to
do that, and see how that pans out. If I had the option, I would prefer
to build my packages in a chroot, test, and then upload only source.
--
see shy jo
Reply to: