rawhide build currently broken

[Florian Weimer]

The i686 build fails with:

gcc   -nostdlib -nostartfiles -r -o 
/builddir/build/BUILD/glibc-2.24-592-g73dfd08/build-i686-redhat-linux/elf/librtld.os 
'-Wl,-(' 
/builddir/build/BUILD/glibc-2.24-592-g73dfd08/build-i686-redhat-linux/elf/dl-allobjs.os 
/builddir/build/BUILD/glibc-2.24-592-g73dfd08/build-i686-redhat-linux/elf/rtld-libc.a 
-lgcc '-Wl,-)' \
	 
-Wl,-Map,/builddir/build/BUILD/glibc-2.24-592-g73dfd08/build-i686-redhat-linux/elf/librtld.os.map
gcc   -nostdlib -nostartfiles -shared -o 
/builddir/build/BUILD/glibc-2.24-592-g73dfd08/build-i686-redhat-linux/elf/ld.so.new 
	\
	  -Wl,-z,combreloc -Wl,-z,relro -Wl,--hash-style=both -Wl,-z,defs 
-Wl,-z,now	\
	 
/builddir/build/BUILD/glibc-2.24-592-g73dfd08/build-i686-redhat-linux/elf/librtld.os 
-Wl,--version-script=/builddir/build/BUILD/glibc-2.24-592-g73dfd08/build-i686-redhat-linux/ld.map

	\
	  -Wl,-soname=ld-linux.so.2			\
	  -Wl,-defsym=_begin=0
/builddir/build/BUILD/glibc-2.24-592-g73dfd08/build-i686-redhat-linux/elf/librtld.os: 
In function `getenv':
/builddir/build/BUILD/glibc-2.24-592-g73dfd08/stdlib/getenv.c:84: 
undefined reference to `strncmp'
collect2: error: ld returned 1 exit status
make[2]: *** [Makefile:460: 
/builddir/build/BUILD/glibc-2.24-592-g73dfd08/build-i686-redhat-linux/elf/ld.so] 
Error 1
make[2]: Leaving directory 
'/builddir/build/BUILD/glibc-2.24-592-g73dfd08/elf'

I can reproduce it upstream with this configure invocation:

../git/configure --prefix=/usr CC="gcc -m32" CFLAGS="-O3 -g" 
CXX="g++ 
-m32" CXXFLAGS="-O3" --enable-tunables --enable-stack-protector=strong 
--build=i686-linux

I believe it was introduced by the tunables changes.

Unfortunately, I probably won't have a chance to fix this until 2017-01-04.

Thanks,
Florian

[Siddhesh Poyarekar]

On Monday 02 January 2017 07:02 PM, Florian Weimer wrote:
>        -Wl,-soname=ld-linux.so.2            \
>        -Wl,-defsym=_begin=0
>  /builddir/build/BUILD/glibc-2.24-592-g73dfd08/build-i686-redhat-linux/elf/librtld.os:
>  In function `getenv':
>  /builddir/build/BUILD/glibc-2.24-592-g73dfd08/stdlib/getenv.c:84:
>  undefined reference to `strncmp'
>  collect2: error: ld returned 1 exit status
>  make[2]: *** [Makefile:460:
>  /builddir/build/BUILD/glibc-2.24-592-g73dfd08/build-i686-redhat-linux/elf/ld.so]
>  Error 1
>  make[2]: Leaving directory
>  '/builddir/build/BUILD/glibc-2.24-592-g73dfd08/elf'
>  
>  I can reproduce it upstream with this configure invocation:
>  
>  ../git/configure --prefix=/usr CC="gcc -m32" CFLAGS="-O3 -g" 
> CXX="g++
>  -m32" CXXFLAGS="-O3" --enable-tunables --enable-stack-protector=strong
>  --build=i686-linux
>  
>  I believe it was introduced by the tunables changes. 
Yes it is in fact due to tunables, where the glibc.malloc.check tunable
is disabled.  I'll test and post a fix tomorrow.

Siddhesh

[Siddhesh Poyarekar]

On Monday 02 January 2017 11:06 PM, Siddhesh Poyarekar wrote:
>  Yes it is in fact due to tunables, where the glibc.malloc.check
> tunable
>  is disabled.  I'll test and post a fix tomorrow. 
Well I couldn't sleep, so here's the patch that fixes this and generally
makes things nicer by eliminating the duplicated strlen and unsetenv
implementations:

https://sourceware.org/ml/libc-alpha/2017-01/msg00065.html

Siddhesh

[Siddhesh Poyarekar]

On Tuesday 03 January 2017 01:09 AM, Siddhesh Poyarekar wrote:
>  Well I couldn't sleep, so here's the patch that fixes this
> and generally
>  makes things nicer by eliminating the duplicated strlen and unsetenv
>  implementations:
>  
>  https://sourceware.org/ml/libc-alpha/2017-01/msg00065.html 
Florian, did you get a chance to review/test my fix?  It works for me
but I want to be sure that things dont break for you.  If you want I can
do a rawhide rebase to verify all architectures - I didn't do it so far
to avoid conflicting with anything that you may have in flight.

Siddhesh

[Carlos O'Donell]

On 01/06/2017 02:55 AM, Siddhesh Poyarekar wrote:
>  On Tuesday 03 January 2017 01:09 AM, Siddhesh Poyarekar wrote:
> > Well I couldn't sleep, so here's the patch that fixes this and generally
> > makes things nicer by eliminating the duplicated strlen and unsetenv
> > implementations:
> >
> > https://sourceware.org/ml/libc-alpha/2017-01/msg00065.html
>  
>  Florian, did you get a chance to review/test my fix?  It works for me
>  but I want to be sure that things dont break for you.  If you want I can
>  do a rawhide rebase to verify all architectures - I didn't do it so far
>  to avoid conflicting with anything that you may have in flight. 
Rawhide i386 build passes with this patch.

Full arch scratch build in progress.

-- 
Cheers,
Carlos.

[Siddhesh Poyarekar]

On Saturday 07 January 2017 02:39 AM, Carlos O'Donell wrote:
>  Rawhide i386 build passes with this patch.
>  
>  Full arch scratch build in progress. 
Thanks!

Siddhesh

[Carlos O'Donell]

On 01/07/2017 12:50 PM, Siddhesh Poyarekar wrote:
>  On Saturday 07 January 2017 02:39 AM, Carlos O'Donell wrote:
> > Rawhide i386 build passes with this patch.
> >
> > Full arch scratch build in progress.
>  
>  Thanks! 
The full build shows some failures related to tunables, but otherwise
fixes the i686 build failure.

I've already commented that upstream.

My next step is to continue building across the RHEL arches and verify
where the failures are coming from.

-- 
Cheers,
Carlos.

[Florian Weimer]

On 01/10/2017 09:59 PM, Carlos O'Donell wrote:
>  On 01/07/2017 12:50 PM, Siddhesh Poyarekar wrote:
> > On Saturday 07 January 2017 02:39 AM, Carlos O'Donell wrote:
> >> Rawhide i386 build passes with this patch.
> >>
> >> Full arch scratch build in progress.
> >
> > Thanks!
>
>  The full build shows some failures related to tunables, but otherwise
>  fixes the i686 build failure.
>
>  I've already commented that upstream.
>
>  My next step is to continue building across the RHEL arches and verify
>  where the failures are coming from. 
Could you push what you currently have to git, or would that be 
premature?  Surely it wouldn't make things worse.

Thanks,
Florian