[uClibc] uClibc 0.9.23 released

Manuel Novoa III mjn3 at codepoet.org
Wed Nov 19 18:52:53 UTC 2003


Hello,

On Wed, Nov 19, 2003 at 09:53:31AM -0700, Greg Haerr wrote:
> :
> : >     I've fixed the install issue, it requires a slightly modified
> : > microwin.mk, attached, and a new microwindows-src-snapshot.tar.gz,
> : > which is on my web site (auto-downloaded).
> : 
> : Builds fine now for i386 but the build fails quickly for arm.
> 
> Manuel,
>     Ok, we'll try round three.  Sorry about sending incomplete software.
> It's been tough to test, because buildroot is still broken in the following
> major areas, at least for me (RH 7.3):

I understand.

> 1. I run a fix.sh to convert "sed -ie" to "perl -p -i -e"

Erik checked in a fix earlier for buildroot to build and use an
updated sed.  I haven't tried it yet.

> 2. No case of USE_UCLIBC_TOOLCHAIN:=false seems
> to work.  The compiler build fails when not building with uclibc.

I'm not surprised.  I honestly don't know why that is there other
than as historical cruft.  Perhaps it was there to allow use of
the gcc wrapper which is no longer supported.

> 3. When specifying the gcc 2.9.5 toolchain, the build immediately
> fails with a non-updated uclibc-configure dependency in gcc-2.95.mk.
> After fixing this, the gcc 2.95 build fails.

The 2.95 build has been broken for quite a while.  I think it was
even broken before my soft float work several weeks ago when I
made a lot of fixes to the 3.2.2 build.  I didn't fix 2.95 because
I don't use it.  Patches welcome.

> 4. The arm gcc build fails unless TARGET_SOFT_FLOAT= option
> is specified.

Sigh... yes.  That's now fixed in cvs.  Erik made some last-minute
chages to the uClibc config system which had unintended side effects.
I checked in fixes yesterday.  You'll need to update your buildroot
sources/uClibc.config* and uClibc/extra/Configs/* files.

> 5. All builds fail trying to run an incorrect-pathed 'ar'.  Typing
> "make" again ignores this step, and the build completes, sometimes.
> 
> So, I've only been able to test this with i386 and arm-nofpu.
> 
> Does anybody actually try to test buildroot before release, or

Well, before I checked in my soft float stuff a couple of weeks ago,
I built normal and soft float versions for i386, arm, and mips, and
mipsel.  That's 8 full buildroot builds.  I also built and ran both
perl 5.8.0 and python 2.3.2 (i think) in the i386 rootfs and ran
their self tests.

> at least run through a couple of important options?  I'm very interested

Depends on what you consider important options.  For instance, I only
recently found out that the OPTIMIZE_FOR_CPU option is broken in the
buildroot Makefile, since I never use it.  Also, the pregenerated
locale stuff only works for x86.  I'll be rewriting much of uClibc
locale support in December.

> in helping make buildroot more stable. Who should I talk to about
> this?

Talk to Erik.

Manuel



More information about the uClibc mailing list