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
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
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
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:
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
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:
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.
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.
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