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

Re: the next step

On Sat Sep 30, 2000 at 06:08:39PM -0700, Joey Hess wrote:
> Daniel Jacobowitz wrote:
> > Oooh, uclibc.  That's a good idea.
> > 
> > In fact, that's a VITAL idea.
> > 
> > Joey, please remember that some architectures (I'm thinking powerpc
> > here, and at least two more) do not support library reduction.  libc on
> > a floppy is not an option.
> I've never been given a good reason for why reduction doesn't work on
> some architectures besides "it just doesn't seem to work". Sigh.

At Lineo (where I work) we have a tool called "Lipo" (written in perl) that
does the library reduction thing, and we have it working quite nicely on a
number of platforms.  In some cases we had to add in special platform specific
lists of symbols to always include, but once you get past the fact that static
analysis sometimes misses sneaky platform specific symbols and just always add
those back in to the keep list, all is well.

> Keep in mind the (single?) floppy install I'm talking about it just a
> demo system, and it's not the only way the new installation system will
> work, just the first one we'll get working.
> So it may be we won't get a single floppy install for some
> architectures. Probably no worse than it is now though.
> I agree uclibc sounds like a good idea. It needs more investigation
> though: does it work on all architectures? Does it support anything
> we'll need to do? Is it painful to use? Is it small enough? (It seems to
> build a 415k libc.a here, which seems a little large.)

I've done a lot of work on uclibc (like the x86 port-in-progress for example).
It currently only supports static linking.  Till now, all supported
architectures have been source tree forks.  We (Lineo/uclinux.org) are
currently in the process of reintegrating all the forks into the one true tree
(using my ever-so-cleaned-up source tree as a base, if I may brag a bit :-)
Currently, I know of at least support for MCORE, ColdFire, DagonBall, 68EN360,
ARM7, SPARC, MIPS, SH2, and x86.  Though I don't quite have x86 ready for heavy
duty production usage (about 30 functions missing still), for many things it
works quite well already. 

I don't think uclibc is quite ready for use by the boot floppies -- but I think
it will be ready before the woody release.  One thing that may need some
attention is arch support (such as Alpha or UltraSparc).


Erik B. Andersen   email:  andersee@debian.org
--This message was written using 73% post-consumer electrons--

Reply to: