[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: sh4 architecture into Wheezy



On Tue, Apr 26, 2011 at 11:29:00PM +0200, Matthias Klose wrote:
> On 04/26/2011 08:36 PM, Aurelien Jarno wrote:
> >On Tue, Apr 26, 2011 at 04:41:23PM +0200, Matthias Klose wrote:
> >>On 04/26/2011 09:39 AM, Neil McGovern wrote:
> >>>I woudn't be particularly happy with that unless the gcc maintainers ok
> >>>it, and I'm still not sure that two days is also an acceptable
> >>>timescale.
> >>
> >>then please drop mips and mipsel as release architectures. At least
> >
> >What is your problem about MIPS? Why do you insist about dropping it? At
> >least be fair and don't spread FUD.
> >
> >GCC on mips/mipsel build in less than 2 days on the recent build
> >machines. It's true that the build time is slightly higher than other
> >architectures, but the testsuite is done on 3 different ABIs. This is
> >something that can be tweaked, as suggested for SH4.
> >
> >Here are the average build time for gcc-4.* since the release of
> >Squeeze [1]:
> >
> >         |  mips  | mipsel |
> >--------+--------+--------+
> >gcc-4.3 |  42864 | 141863 |
> >gcc-4.4 | 104400 | 149148 |
> >gcc-4.5 | 123498 | 114435 |
> >gcc-4.6 |  95725 | 167799 |
> 
> gcc-4.6: 167799/3600 = 46.61, and this is with the libstdc++
> testsuite already disabled, because it did timeout or fail on the
> mipsel buildds. So this is *no* FUD.  Did you look at the build

So you want to drop both mips and mipsel because mipsel is slow? Great.

> failures, or some other mips porter, before I did disable the tests?

Oh you disabled the libstdc++ testsuite? What about asking/warning the
porters next time?

> >The build time dispersion is explained by the fact we have buildds of
> >different speed, gcc-* is built by default on them (no_weak_autobuild),
> >unless this build daemon is already busy.
> >
> >
> >>sh4 has a workable, accessible developer machine,
> >
> >mips also has an accessible developer machine, gabrielli.debian.org.
> >It's true that mipsel doesn't have one (it's being working on), that
> >said, most issues are reproducible on both. People can also ask on
> >debian-mips for help in case it's a mipsel specific issue.
> >
> >
> >>and people within
> >>Debian who care about the architecture.
> >
> >MIPS also has Debian people who care about the architecture. See for
> >example my recent MIPS work:
> >
> >http://svn.debian.org/wsvn/kernel/?op=comp&compare%5B%5D=%2Fdists%2Fsid%2Flinux-2.6%2Fdebian@17159&compare%5B%5D=%2Fdists%2Fsid%2Flinux-2.6%2Fdebian@17161
> >http://svn.debian.org/wsvn/gcccvs/?op=comp&compare%5B%5D=%2Fbranches%2Fsid%2Fgcc-4.6%2Fdebian@5248&compare%5B%5D=%2Fbranches%2Fsid%2Fgcc-4.6%2Fdebian@5262
> >http://svn.debian.org/wsvn/gcccvs/?op=comp&compare%5B%5D=%2Fbranches%2Fsid%2Fgcc-4.5%2Fdebian@5263&compare%5B%5D=%2Fbranches%2Fsid%2Fgcc-4.5%2Fdebian@5267
> >http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=623014
> >http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=623015
> >http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=623162
> >http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=623598
> >http://lists.debian.org/debian-mips/2011/04/msg00003.html
> >http://lists.debian.org/debian-mips/2011/04/msg00018.html
> >http://sourceware.org/bugzilla/show_bug.cgi?id=12606
> 
> yes, the last one incomplete and only completed by myself.  So who

So basically reporting an identical bug in bugzilla without checking 
if a bug already exists is completing?

> else is doing toolchain work on mips in Debian?  Thiemo did leave a
> big gap, and it was an effort of many people to release squeeze with
> mips.  I just see that

I usually report bug upstream when needed (sometimes even providing
patches), and follow the upstream development to enable new features
(e.g.: SSP support, -mplt, defaulting to MIPS II ABI). I agree I 
sometimes lack of time, but I don't see much more work than that on
other architectures.

> >All that said, I agree that mips and mipsel architectures are not in
> >their best shape, but people are working on that. If you consider they
> >don't follow the release criteria, please give objective arguments.
> 
> the build time argument was brought up by the debian-release team,
> so this this seems to be an objective argument. If not, maybe the
> release criteria for new, current and "obsolet" ports should be made
> more transparent. I'm only aware of one table not differentiating
> new and current ports.

The SH4 people announced that they don't have any faster hardware and 
that GCC needs 2 days with all the tests disabled.

It's clearly not comparable with mipsel which needs, with some of the
tests being run, slightly more than 2 days when built on the slowest 
buildd, and about 1 day 6h on the fastest one, and for which
we are waiting for faster hardware.

> yes, other issues are the non-availabilty of a mipsel porter box and
> the instability of the existing mips porter box.
> 
> and toolchain maintenance was rather difficult (longsoon, binutils)
> during the squeeze cycle.

The binutils issue was a real toolchain issue. As for the longsoon part,
it is a silicon issue, not a toolchain one. It was workarounded by one
patch in binutils and one patch in gcc. How can you consider that part
difficult toolchain maintenance?

>   Matthias
> 
> Please note that this thread did start about sh4, and some comments
> about the sh4 toolchain by some members of the release team, which
> apply for mips* too, and which are used against the sh4 port.

Yes, and as always when you can hijack a thread or an IRC discussion to
complain about mips, you don't hesitate.

> I appreciate your work on mips, but I think a lot more needs to be
> done to keep it as a release architecture, and that arguments that
> are overlooked by intent for existing release architectures should
> not be used against a new port.

The release team has clear release criteria, but allow a few of them
to not be fully satisfied, otherwise no other architecture than amd64
and i386 would stay in the archive. In any case criteria have to be 
compared objectively.

-- 
Aurelien Jarno	                        GPG: 1024D/F1BCDB73
aurelien@aurel32.net                 http://www.aurel32.net

Attachment: signature.asc
Description: Digital signature


Reply to: