Re: ppc64le test build failure on glibc 2.23.

["Carlos O'Donell" <carlos at redhat dot com>]

On 02/18/2016 10:32 PM, Carlos Eduardo Seo wrote:
> Looking at the preprocessor output for test-get_hwcap, I see it's getting the header from the system:
> 
> Fedora 21:
> 
> # 1 "/usr/include/sys/auxv.h" 1 3 4
> # 20 "/usr/include/sys/auxv.h" 3 4
> 
> Ubuntu 14.04:
> 
> # 1 "/usr/include/powerpc64le-linux-gnu/sys/auxv.h" 1 3 4
> # 20 "/usr/include/powerpc64le-linux-gnu/sys/auxv.h" 3 4
> 
> I suppose we can't always rely on this and it has to be fixed for 2.23. What solution do you prefer?

That's mixing host glibc headers while testing new glibc.
That's a recipe for problems.

For good testing purposes you should isolate your builds as much
as possible from the host using --with-headers, and it's because
of this that I see the problem. When using --with-headers you have
to build up a sysroot include tree to use with the build (adds
-nostdinc to builds).

Have a look at the scripts I'm attaching for how I do glibc builds
(needs kernel sources to get UAPI headers, but it does it all
automatically).

My suggested fix is to add auxv.h wrapper to include/sys/, and fixup
elf/test-auxv.c, and then see if there is any other fallout.

Cheers,
Carlos.

Attachment: build-glibc.sh
Description: application/shellscript

Attachment: glibc-config.sh
Description: application/shellscript