Re: Legal problems of unbuildable testing (Was: ... towards Debian 3.1)
goswin wrote:
> aj wrote:
> > If we ship a new compiler that does optimisations slightly differently
> > to the previous one, the binaries we built previously won't match the
> > new compiler either. This isn't a reasonable argument, and wasting time
> > worrying about it doesn't benefit our users even slightly.
And if the new compiler has a rarely-triggered optimisation bug, and a
package built with an old compiler has to be updated in stable because
of a security bug, then the workload on the people doing the security
fix dramatically increases, at a time where things ought to be done
damn fast...
This is all about software reproducibility, something that all serious
software vendors have to handle if they don't want to loose time
uselessly to support their products (and yes, I've had real-life
experience with that).
To do things right, we'd have to freeze the toolchain early, and
ensure everything in a release has been built with the proper
toolchain.
> GPL does not say you have to ship the build environment with the
> source.
I think it could be argued that the build environment is part of the
"system libraries and build scripts" stuff, but the point can be made
without that.
> You don't have to ship the system libraries or sources for the same
> with it either.
As I said, since we are the one shipping those "system libraries",
it'd be ridiculous to use this clause to justify it when we ship a
program built with an old version of a lib, together with a newer
version of that lib, with which that program cannot be made to build.
Right ?
Now I'm not saying we have done that, but I'm not saying we haven't
either. But the large number of "FTBFS on i386" which I have myself
received on my own packages during the last year makes me think that
we have most probably dones that in past releases.
The i386 buildd's used to track such build problems can not detect
everything - like features added or suppressed from a lib, whose use
would depend on the version of the lib's headers used to build the
program.
And well, I guess we could find an endless list of such problems. Why
not instead fix things the Right Way, and get our releases self-built !?
> Its perfectly fine to shipd a binary thats compiled by a non-free
> prorietary compiler noone else has access to. No problems there.
OK, but not true for Debian.
Regards,
--
Yann Dirson <Yann.Dirson@fr.alcove.com> http://www.alcove.com/
Technical support manager Responsable de l'assistance technique
Senior Free-Software Consultant Consultant senior en Logiciels Libres
Debian developer (dirson@debian.org) Développeur Debian
Reply to: