ERRNO handling in linuxthreads.old (broken build)(not solved)

[Sergei Poselenov]

Hello Bernhard,

On Чт., 2013-03-21 at 19:50 +0100, Bernhard Reutner-Fischer wrote:
> On 19 March 2013 16:47, Sergei Poselenov <sposelenov at emcraft.com> wrote:
> > Hello all,
> >
> > As it turned out, uncommenting the "!STATIC" condition in
> > include/libc-symbols.h breaks the "libc_hidden_proto()" for config_read().
> > I'm building busybox with its own version of config_read() and resolver gets
> > linked with the busybox version. It's the same issue as
> > http://lists.uclibc.org/pipermail/uclibc/2010-November/044488.html
> >
> > Anyone willing to help? How can build linuxthreads.old in static configuration
> > so that the errno could be correctly initialized?
> 
> Can you reproduce the problem on master?
> If not,
> cherry-pick 87936cd0130414cccc8a7427498dd5491ec2664f and
> f418f52701de02954c90341e5c9ccdd3f07ccd33 unto the 0.9.33 branch ?

I tried but master (with my patches for thumb/testandset()) doesn't work
for me. A simple MT application just exits after ptrhead_create(). The
same with the 0.9.33 (patched) with these two commits cherry-picked.
Plain 0.9.33 (patched) creates the thread but it has the original
problem referencing to the correct errno location.

Anyway, thanks for the hint.

I finally found and applied to 0.9.32 the patch by Peter Korsgaard
https://github.com/twobob/buildroot-kindle/blob/master/toolchain/uClibc/uClibc-0.9.31.1-linuxthreads-errno-fix.patch.

With the patch applied all works as expected. My tests included:
 - errno location is correct for the thread
 - busybox nslookup simple test to ensure config_read() linked to
resolver is of the libc version, not from busybox (otherwise SEGV-ed).

 As I learned, the patch wasn't applied to uClibc master because it
breaks NPTL build on SPARC, correct?

And BTW when do you expect 0.9.33 released? 

Regards,
Sergei