On Fri, Apr 29, 2011 at 10:53:18PM +0200, Didier Raboud wrote: > Le vendredi, 29 avril 2011 22.17:51, Stephen Kitt a écrit : > > There's no technical limitation, although I haven't tried such > > cross-builds. I limited the architectures following a comment on the > > gcc-mingw-w64 ITP (see > > http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=602997#14 and my > > reply just below that), although Matthias was perhaps more concerned > > with excessive build times on some architectures. So kfreebsd-amd64, > > powerpc etc. would probably make sense - I'll check the gcc build > > times on the various architectures. > > IMHO, this limitation is useless: it's not because we might think that > building Windows executables on non-mainstream architectures (or OSes fwiw) > that anyone should be forbidden to try.
Indeed, but it comes at a significant cost on some of the buildds.
Given that the request not to specify “Architecture: any” on
gcc-mingw-w64 comes from one of the gcc maintainers who has to deal
with the fallout from lengthy build times fairly regularly, I'm
inclined to respect it even though the resulting limitation is
artificial (from an end-user's perspective).
The only technical counter-argument I can see is that of potential
reverse build-dependencies; for example nsis supports more
architectures than {binutils,gcc}-mingw-w64, which is unfortunate. (It
doesn't support arm though; the next version of nsis will switch to
“Architecture: all” packages for Win32-specific files, which will
solve the problem.)
Thinking specifically of this bug against binutils-mingw-w64, build
time is much less of a problem, so specifying “Architecture: any”
would be OK there. I don't know if many people would find
binutils-mingw-w64 useful without gcc-mingw-w64 but there might be
some!
> I think that an easy way to determine if it can be worth would be to upload a
> 0.1+exp1 package with s/i386 amd64/any/ to experimental. Buildds will do this
> job for you for free.
From the gcc-4.6 build logs:
* armel: 22h
* hurd-i386: 6h35
* i386: 5h30
* ia64: failed
* kfreebsd-amd64: 12h50
* kfreebsd-i386: 10h
* mips: 25h35
* mipsel: still building after over 30h
* powerpc: 11h40
* s390: 6h
* sparc: 18h40
> What do you think ?
I reckon it would make sense to do the following:
* change binutils-mingw-w64 to “Architecture: any” (thus closing this
bug);
* change gcc-mingw-w64 to “Architecture: amd64 hurd-any i386
kfreebsd-any powerpc s390 sparc” for now.
Would that be good enough for you?
Regards,
Stephen
signature.asc
Description: Digital signature
