Sync gnustep-base 1.24.6-2 (universe) from Debian unstable (main)

This bug was fixed in the package gnustep-base - 1.24.6-2
Sponsored for Paul Gevers (paul-climbing)

---------------
gnustep-base (1.24.6-2) unstable; urgency=medium

  * Upload to unstable.
  * debian/compat: Set to 9 to get .build-id debugging symbols.
  * debian/control.m4 (Build-Depends): Require debhelper >= 9. Build
    against gnutls28, thanks Andreas Metzler (Closes: #753033).
    (libgnustep-base`'SOV_BASE-dbg) <Conflicts>: Remove.
    <Recommends>: Remove libobjc3-dbg.
  * debian/control: Regenerate.
  * debian/patches/hurd-ignore-NSURL-test.diff: Delete; apparently no
    longer needed.
  * debian/patches/ftbfs-mips64.patch: New; fixes FTBFS on mips64(el),
    thanks Yunqiang Su (Closes: #753961).
  * debian/patches/gnutls28.patch: New; avoid using and linking libgcrypt
    with recent GnuTLS.
  * debian/patches/NSString-buffer-overrun.patch: New, cherry-picked from
    upstream SVN.
  * debian/patches/alignment-check.patch: New, fix word-alignment
    configure test on sparc. Cherry-picked from upstream.
  * debian/patches/testsuite-fixes.patch: New.
  * debian/patches/doc-links.patch: Point links to index.html.
  * debian/patches/autoreconf.patch: Regenerate.
  * debian/patches/series: Update.
  * debian/rules (confflags): Define conditionally and pass --host to
    configure if cross-compiling.
    (install-debug): Install config.log and tests.log to analyze testsuite
    failures on the buildds.

 -- Yavor Doganov <email address hidden> Mon, 07 Jul 2014 12:16:59 +0300

gnustep-base (1.24.6-1) experimental; urgency=low

  * New upstream release:
    - Fixes FTBFS with recent libxml2 (Closes: #738347).
    - GNUSTEP_USER_DIRECTORY is no longer created unconditionally (Closes:
      #720190).
    - Fixes regression in performSelector: with message forwarding
      (Closes: #753603).
  * Ack NMUs; thanks Matthias Klose, gregor herrmann and Pino Toscano.
  * debian/patches/libobjc4.patch:
  * debian/patches/recent-libxml2-fix.patch: Remove; fixed upstream.
  * debian/patches/kfreebsd-fake-main.patch:
  * debian/patches/avoid-nsl-linkage.patch:
  * debian/patches/maxsymlinks.diff: Refresh.
  * debian/patches/autoreconf.patch: Regenerate.
  * debian/patches/texinfo5.diff: Add description.
  * debian/patches/hurd-ignore-NSURL-test.diff: Disable for now.
  * debian/patches/manpage-fixes.patch: Fix two more issues reported by
    lintian.
  * debian/patches/info-direntry.patch: Fix few texinfo warnings.
  * debian/patches/CVE-2014-2980.patch: New patch from upstream, fixes
    gdomap user security hole (Closes: #745470).
  * debian/patches/use-local-DTDs.patch: New; use local DTDs to avoid
    annoying warnings from autogsdoc when built in a chroot. Thanks
    Svante Signell (Closes: #736587).
  * debian/patches/hide-SYSTEM_CONFIG-vars.patch: New; fix for upstream
    bug #42423.
  * debian/patches/doc-links.patch: New; fix some broken links to manuals
    in the various -doc packages, thanks js (Closes: #749196).
  * debian/patches/series: Update.
  * debian/rules (build-arch): Remove dependency on patch.
    (binary-indep): Invoke dh_installxmlcatalogs with -n since only DTDs
    are being installed, not catalogs (Closes: #637093).
    (i...

Hi Yavor,

gnustep-base is failing in Ubuntu on powerpc and ppc64el while it worked
in the past. The error in both builds is following. Do you have any idea
why objc would be not working for gnustep-base? Would this be a bug in
objc or should we (I) fix something for gnustep-base.

checking whether objc really works... no
I don't seem to be able to use your Objective-C compiler to produce
working binaries! Please check your Objective-C compiler installation.
If you are using gcc-3.x make sure that your compiler's libgcc_s and libobjc
can be found by the dynamic linker - usually that requires you to play
with LD_LIBRARY_PATH or /etc/ld.so.conf.
Please refer to your compiler installation instructions for more help.
configure: error: The Objective-C compiler does not work or is not
installed properly.

Full logs can be found here:
https://launchpadlibrarian.net/180283788/buildlog_ubuntu-utopic-powerpc.gnustep-base_1.24.6-2_FAILEDTOBUILD.txt.gz
https://launchpadlibrarian.net/180283768/buildlog_ubuntu-utopic-ppc64el.gnustep-base_1.24.6-2_FAILEDTOBUILD.txt.gz

Paul

-------- Original Message --------
Subject: [Bug 1343945] Re: Sync gnustep-base 1.24.6-2 (universe) from
Debian unstable (main)
Date: Sat, 19 Jul 2014 09:04:38 -0000
From: Artur Rona <email address hidden>
Reply-To: Bug 1343945 <email address hidden>
To: <email address hidden>

This bug was fixed in the package gnustep-base - 1.24.6-2
Sponsored for Paul Gevers (paul-climbing)

On 19-07-14 11:49, Yavor Doganov wrote:
> Paul Gevers wrote:
>> gnustep-base is failing in Ubuntu on powerpc and ppc64el while it
>> worked in the past.
>
> Very unfortunate. Does it build on all other architecures?

No, only succeeded on arm (64 and hf).

>> The error in both builds is following. Do you have any idea why objc
>> would be not working for gnustep-base? Would this be a bug in objc
>> or should we (I) fix something for gnustep-base.
>
> I need config.log to analyze this. Can you obtain it?

I have no idea, but I expect not. I think will create a Ubuntu chroot
and build myself (on amd64) so I can add it here. I will get back to
you, but if somebody can get to the files before I can create them, much
appreciated.

Paul

Hi,

On 19-07-14 13:00, Paul Gevers wrote:
> On 19-07-14 11:49, Yavor Doganov wrote:
>> Paul Gevers wrote:
>>> gnustep-base is failing in Ubuntu on powerpc and ppc64el while it
>>> worked in the past.
>>
>> Very unfortunate. Does it build on all other architecures?
>
> No, only succeeded on arm (64 and hf).

But the build "error" is different on amd64. It times out in one of the
testcases.
--- Running tests in base/NSOperation ---

Session terminated, terminating
shell.../usr/share/GNUstep/Makefiles/Master/serial-subdirectories.make:53:
recipe for target 'internal-check' failed
make[1]: *** [internal-check] Terminated
debian/rules:140: recipe for target 'debian/build-shared-stamp' failed
make: *** [debian/build-shared-stamp] Terminated
GNUmakefile:61: recipe for target 'check' failed
make[2]: *** [check] Terminated
 ...terminated.
Build killed with signal 15 after 60 minutes of inactivity

>>> The error in both builds is following. Do you have any idea why objc
>>> would be not working for gnustep-base? Would this be a bug in objc
>>> or should we (I) fix something for gnustep-base.
>>
>> I need config.log to analyze this. Can you obtain it?
>
> I have no idea, but I expect not. I think will create a Ubuntu chroot
> and build myself (on amd64) so I can add it here.

And of course it succeeds in my chroot. So it seems something specific
for the build servers. Actually, the current behavior was already
reported in [1].

https://bugs.launchpad.net/ubuntu/+source/gnustep-base/+bug/1277975

I propose to build without the tests until the underlaying problem is
fixed. Any objections?

Paul

Oops, but that doesn't fix the powerpc/ppc64el failures. I am afraid I
don't know how to get these logs... maybe have them written to stdout
before the failure so that they show up in the build logs?

Paul

On 19-07-14 15:10, Paul Gevers wrote:
> Hi,
>
> On 19-07-14 13:00, Paul Gevers wrote:
>> On 19-07-14 11:49, Yavor Doganov wrote:
>>> Paul Gevers wrote:
>>>> gnustep-base is failing in Ubuntu on powerpc and ppc64el while it
>>>> worked in the past.
>>>
>>> Very unfortunate. Does it build on all other architecures?
>>
>> No, only succeeded on arm (64 and hf).
>
> But the build "error" is different on amd64. It times out in one of the
> testcases.
> --- Running tests in base/NSOperation ---
>
> Session terminated, terminating
> shell.../usr/share/GNUstep/Makefiles/Master/serial-subdirectories.make:53:
> recipe for target 'internal-check' failed
> make[1]: *** [internal-check] Terminated
> debian/rules:140: recipe for target 'debian/build-shared-stamp' failed
> make: *** [debian/build-shared-stamp] Terminated
> GNUmakefile:61: recipe for target 'check' failed
> make[2]: *** [check] Terminated
> ...terminated.
> Build killed with signal 15 after 60 minutes of inactivity
>
>>>> The error in both builds is following. Do you have any idea why objc
>>>> would be not working for gnustep-base? Would this be a bug in objc
>>>> or should we (I) fix something for gnustep-base.
>>>
>>> I need config.log to analyze this. Can you obtain it?
>>
>> I have no idea, but I expect not. I think will create a Ubuntu chroot
>> and build myself (on amd64) so I can add it here.
>
> And of course it succeeds in my chroot. So it seems something specific
> for the build servers. Actually, the current behavior was already
> reported in [1].
>
> https://bugs.launchpad.net/ubuntu/+source/gnustep-base/+bug/1277975
>
> I propose to build without the tests until the underlaying problem is
> fixed. Any objections?
>
> Paul
>

On 20-07-14 17:16, Yavor Doganov wrote:
> Yes, this is one way of doing it. Something like this
>
> - $(confflags)
> + $(confflags) || (cat config.log; exit 1)
>
> in debian/rules (configure-stamp) ought to work.

Yes, let's do this. I propose to add this to the Debian package as well.
I will create a debdiff to do this in Ubuntu now.

>>> Build killed with signal 15 after 60 minutes of inactivity
>
> This is very similar to Debian #649419. Unfortunately I was not able
> to reproduce it after probably 100 builds of gnustep-base. You have
> to attach gdb to the stuck test program and see what's going on.

Well, a retry succeeded. So it is not something that we need to fix on
short term. I think someone with access to a build server setup should
investigate this.

Paul

Forwarded for the bug tracker.

-------- Original Message --------
Subject: Re: Fwd: [Bug 1343945] Re: Sync gnustep-base 1.24.6-2
(universe) from Debian unstable (main)
Date: Wed, 23 Jul 2014 10:02:53 +0300
From: Yavor Doganov <email address hidden>
To: Bug 1343945 <email address hidden>

(Paul, CCing you because for some reason my messages don't end up in
the Ubuntu bug log.)

From the build logs:

| /usr/lib/gcc/powerpc-linux-gnu/4.8/libobjc.so: undefined reference to
| `rs6000_special_adjust_field_align_p'

This is apparently due to debian/patches/gcc-elfv2-abi-warn3.diff.
This patch has not been committed to GCC upstream AFAICS although it
has been ACKed. It's very difficult for me to judge about the
implications; perhaps the patch is incomplete as linking fails.

It was also added to Debian's gcc-4.8 in 4.8.3-5 so will probably fail
on Debian, too (if gnustep-base is built with gcc-4.8).

I don't have any ideas how to solve this other than disabling the
patch. This, however, is probably not an option as there may have
been good reasons to include it. Someone familiar with GCC/powerpc
should take a look and comment... Another option is to build with
gcc-4.9, which will require changes to gnustep-make if it is applied
for all GNUstep packages. Again, this might be unacceptable.

Out of curiosity, I couldn't find the reason why gcc-4.8 is the
default compiler in Ubuntu. That's very odd; usually Ubuntu is ahead
of Debian wrt toolchain updates.

Forwarded for the bug tracker.

-------- Original Message --------
Subject: Re: Fwd: [Bug 1343945] Re: Sync gnustep-base 1.24.6-2
(universe) from Debian unstable (main)
Date: Wed, 23 Jul 2014 10:34:12 +0300
From: Yavor Doganov <email address hidden>
To: Bug 1343945 <email address hidden>

Yavor Doganov wrote:
> | /usr/lib/gcc/powerpc-linux-gnu/4.8/libobjc.so: undefined reference to
> | `rs6000_special_adjust_field_align_p'
>
> It's very difficult for me to judge about the implications; perhaps
> the patch is incomplete as linking fails.

Looking at the gcc-4.8 build log, there are warnings when building
libobjc:

/«PKGBUILDDIR»/src/libobjc/encoding.c: In function
'objc_layout_structure_next_member':
/«PKGBUILDDIR»/src/libobjc/encoding.c:1163:3: warning: implicit
declaration of function 'rs6000_special_adjust_field_align_p'
[-Wimplicit-function-declaration]
   desired_align = ADJUST_FIELD_ALIGN (type, desired_align);
   ^

It seems that an #include is missing as the function prototype is not
available. So my conclusion that the patch is incomplete seems to be
correct. Anything that links with libobjc on powerpc will fail.

On 23-07-14 09:02, Yavor Doganov wrote:
> Another option is to build with
> gcc-4.9, which will require changes to gnustep-make if it is applied
> for all GNUstep packages. Again, this might be unacceptable.

Can you be more specific? Do we need to build everything again starting
from gnustep-make if we want all gnustep packages to build with gcc-4.9?
It seems that most recent imports of gnustep packages into Ubuntu all
fail to build. If you/we believe that gnustep should currently be build
with (at least) 4.9 we just have to put that in the build-depends (also
in Debian if we can). I am wondering, for gnustep to build properly, do
all the gnustep packages need to be build with the same gcc version?
Maybe the problem currently is a mixing of versions.

> Out of curiosity, I couldn't find the reason why gcc-4.8 is the
> default compiler in Ubuntu. That's very odd; usually Ubuntu is ahead
> of Debian wrt toolchain updates.

I think it is this:
gcc-defaults (1.128ubuntu4) utopic; urgency=medium

  * Revert default to 4.8, due to as-yet-undiagnosed C++ ABI breakage
    (LP: #1329089).

 -- Steve Langasek <email address hidden> Wed, 11 Jun 2014
23:31:26 +000

Paul