Re: Compiling Debian from source, with GCC 3.4.2
Martin Michlmayr wrote:
> * Laurence Darby <ldarby@mips.com> [2005-02-28 21:31]:
> > It's my task to compile a complete Debian distribution from source.
> > The main goal is to test our version of the GCC compiler, SDE, which
> > is based on 3.4.2, and has more tuning/optimisations for the MIPS
> > architecture
>
> Are you planning to do the same work with GCC 4.1? If not, I might
> give it a go, but I wouldn't mind if someone else would do it... ;-)
Recompiling Debian isn't on my task list anymore, see below for why.
But if you want to, go ahead. I doubt any of the scripts I wrote
would be useful, as they were in tcsh and pretty kludgy. In hindsight
this was a case of NIH, I should have started with buildd, or was it
sbuild, or was it wanna-build, or was it wtf-do-you-build debian with...
I hope for your sake the builds don't fail because of things like
$ gcc --version
not outputing the string "gcc-3.3"
> Also, how did your effort to compile the archive with MIPS' SDE work
> out in the end?
Not that well. I found a couple of GCC bugs, after compiling over
12000 deb packages, and about 400 failures that didn't look like SDE or
MIPS specific compiler bugs (broken build-deps etc). That was about as
far as I got before moving on to other projects.
The main problem was that I do not have the endurance to follow and
debug the instructions to create a mirror out of the the 12000 packages
(or even just the 100 base packages), then to build a debian-installer
that used these packages. The experience left me wanting to post the
entire Debian project to thedailywtf.com, because I was going "WTF??"
on a daily basis.
I guess you need a PhD to understand Debian, I've got only a degree.
(Yes, Martin, I've read your bio...)
The other major problem, ie. demotivation, was that even though it's
real world code for the compiler to chew on, it wasn't testing coverage
of the options (-O3 -finline-loops -fwhatever) , and there's no
sustainable and procedural way to test all the generated code. That's
what is needed for verification as well as not segfaulting on random C
code.
Laurence
Reply to: