build bot failure

build bot failure

[Jim Wilson]

I got a message from the build bot after checking in a binutils patch,
but I don't see how my patch can relate to the build bot failure.

ld: 0711-380 STABSTRING ERROR: Symbol table entry 534, object file
../bfd/libbfd.a[compress.o]
Length of stabstring in .debug section is invalid.
The stabstring is being deleted.

It looks like the AIX linker is objecting to debug info emitted by
gcc, and I didn't touch either of those things.  I only touched
binutils.  I don't have access to an AIX system, so I am not able to
reproduce the problem and debug it.  I am able to build gdb on an
x86-linux and an aarch64-linux system.  I suspect a problem on the aix
build slave, but don't know how to check that.  I'm not sure what to
do next.

Jim

Re: build bot failure

[Yao Qi]

Jim Wilson <jim.wilson@linaro.org> writes:

> I got a message from the build bot after checking in a binutils patch,
> but I don't see how my patch can relate to the build bot failure.
>
> ld: 0711-380 STABSTRING ERROR: Symbol table entry 534, object file
> ../bfd/libbfd.a[compress.o]
> Length of stabstring in .debug section is invalid.
> The stabstring is being deleted.

FYI, the full log can be found
http://gdb-build.sergiodj.net/builders/AIX-POWER7-plain/builds/2359/steps/compile%20gdb/logs/stdio

>
> It looks like the AIX linker is objecting to debug info emitted by
> gcc, and I didn't touch either of those things.  I only touched
> binutils.  I don't have access to an AIX system, so I am not able to
> reproduce the problem and debug it.  I am able to build gdb on an
> x86-linux and an aarch64-linux system.  I suspect a problem on the aix
> build slave, but don't know how to check that.  I'm not sure what to
> do next.

The AIX slave is gcc111.fsffrance.org, from gcc compile farm.  If you
have gcc compile farm account, you can access to it.

-- 
Yao (齐尧)

Re: build bot failure

[Yao Qi]

Yao Qi <qiyaoltc@gmail.com> writes:

> The AIX slave is gcc111.fsffrance.org, from gcc compile farm.  If you
> have gcc compile farm account, you can access to it.

I can reproduce this error on the AIX slave.  Note that you need
--disable-werror in configure and some tweaks on missing makeinfo in bfd
to get the final linking step.

-- 
Yao (齐尧)

Re: build bot failure

[Jim Wilson]

On Tue, May 24, 2016 at 3:38 AM, Yao Qi <qiyaoltc@gmail.com> wrote:
> Yao Qi <qiyaoltc@gmail.com> writes:
>
>> The AIX slave is gcc111.fsffrance.org, from gcc compile farm.  If you
>> have gcc compile farm account, you can access to it.

I requested a gcc compile farm account so I can try to reproduce.

Jim

Re: build bot failure

[Pedro Alves]

On 05/24/2016 11:38 AM, Yao Qi wrote:
> Yao Qi <qiyaoltc@gmail.com> writes:
> 
>> The AIX slave is gcc111.fsffrance.org, from gcc compile farm.  If you
>> have gcc compile farm account, you can access to it.
> 
> I can reproduce this error on the AIX slave.  Note that you need
> --disable-werror in configure and some tweaks on missing makeinfo in bfd
> to get the final linking step.

Did you try the commit before Jim's?  Maybe would helps determine whether
this is a regression caused by Jim's patch, or some other issue with
the slave's toolchain.

Thanks,
Pedro Alves

Re: build bot failure

[Yao Qi]

Pedro Alves <palves@redhat.com> writes:

> Did you try the commit before Jim's?  Maybe would helps determine whether
> this is a regression caused by Jim's patch, or some other issue with
> the slave's toolchain.

Build is OK with the commit before Jim's.

-- 
Yao (齐尧)

Re: build bot failure

[Jim Wilson]

On Wed, May 25, 2016 at 1:07 AM, Yao Qi <qiyaoltc@gmail.com> wrote:
> Pedro Alves <palves@redhat.com> writes:
>
>> Did you try the commit before Jim's?  Maybe would helps determine whether
>> this is a regression caused by Jim's patch, or some other issue with
>> the slave's toolchain.
>
> Build is OK with the commit before Jim's.

I haven't forgotten about this.  I sent email asking for a compile
farm account 2.5 weeks ago and got no response.  I've pinged that
mesage.  I see that this is still broken in the build bot, but I won't
be able to look at it until I get a compile farm account.

Jim

Re: build bot failure

[Sergio Durigan Junior]

On Friday, June 10 2016, Jim Wilson wrote:

> On Wed, May 25, 2016 at 1:07 AM, Yao Qi <qiyaoltc@gmail.com> wrote:
>> Pedro Alves <palves@redhat.com> writes:
>>
>>> Did you try the commit before Jim's?  Maybe would helps determine whether
>>> this is a regression caused by Jim's patch, or some other issue with
>>> the slave's toolchain.
>>
>> Build is OK with the commit before Jim's.
>
> I haven't forgotten about this.  I sent email asking for a compile
> farm account 2.5 weeks ago and got no response.  I've pinged that
> mesage.  I see that this is still broken in the build bot, but I won't
> be able to look at it until I get a compile farm account.

FWIW, I have just disabled e-mail notifications for the AIX machine (and
also for the RHEL-7.1 s390x).  Their reports do not seem to be reliable,
so I don't think it's good to issue the warnings and waste time of the
developers.

I am Cc'ing David Edelsohn, who is the maintainer of both machines.
David, if you could take a look at them and fix these errors it'd be
awesome.

Thanks,

-- 
Sergio
GPG key ID: 237A 54B1 0287 28BF 00EF  31F4 D0EB 7628 65FC 5E36
Please send encrypted e-mail if possible
http://sergiodj.net/

Re: build bot failure

[David Edelsohn]

On Fri, Jun 10, 2016 at 2:47 PM, Sergio Durigan Junior
<sergiodj@redhat.com> wrote:
> On Friday, June 10 2016, Jim Wilson wrote:
>
>> On Wed, May 25, 2016 at 1:07 AM, Yao Qi <qiyaoltc@gmail.com> wrote:
>>> Pedro Alves <palves@redhat.com> writes:
>>>
>>>> Did you try the commit before Jim's?  Maybe would helps determine whether
>>>> this is a regression caused by Jim's patch, or some other issue with
>>>> the slave's toolchain.
>>>
>>> Build is OK with the commit before Jim's.
>>
>> I haven't forgotten about this.  I sent email asking for a compile
>> farm account 2.5 weeks ago and got no response.  I've pinged that
>> mesage.  I see that this is still broken in the build bot, but I won't
>> be able to look at it until I get a compile farm account.
>
> FWIW, I have just disabled e-mail notifications for the AIX machine (and
> also for the RHEL-7.1 s390x).  Their reports do not seem to be reliable,
> so I don't think it's good to issue the warnings and waste time of the
> developers.
>
> I am Cc'ing David Edelsohn, who is the maintainer of both machines.
> David, if you could take a look at them and fix these errors it'd be
> awesome.

I had been testing GCC 6.1 on the RHEL s390x system and the buildbot
apparently picked up that compiler, which generated additional
warnings that were treated as errors. GCC 6 apparently finds more
warnings in GDB, which needs to be addressed in GDB.

The AIX system has not changed, so some change to GDB broke the build
on AIX.  It's warning about missing declarations. Is this related to
the C++ change?

Thanks, David

Re: build bot failure

[Jim Wilson]

On Sat, Jun 11, 2016 at 5:26 AM, David Edelsohn <dje.gcc@gmail.com> wrote:
> The AIX system has not changed, so some change to GDB broke the build
> on AIX.  It's warning about missing declarations. Is this related to
> the C++ change?

An aarch64 binutils patch from me broke it.  I can't see anything
wrong with my patch.  I can't tell why AIX builds are broken without
access to an AIX system, and my request for a compile farm account 2.5
weeks ago hasn't gotten a response yet, so I'm stuck.

https://sourceware.org/ml/binutils/2016-05/msg00103.html
https://sourceware.org/ml/gdb-testers/2016-q2/msg03281.html

Jim

Re: build bot failure

[David Edelsohn]

Jim,

The problem is the debug stabstring for bfd_reloc_code_real_names now
overflows the maximum length.  Your patch overflowed the limit.

More recent releases of GCC for AIX added support for stab
continuation strings, but not the release installed on the machine
running the buildbot.

There was a similar problem in GCC about a year ago.  Where it is
stabs or DWARF, a single debug statement that large implies something
is wrong with the design of that data structure.

Also, if you still want an account on the AIX system in the GNU
Compile Farm, send me your public key, I can create it while we wait
for Laurent.

- David


On Sat, Jun 11, 2016 at 1:00 PM, Jim Wilson <jim.wilson@linaro.org> wrote:
> On Sat, Jun 11, 2016 at 5:26 AM, David Edelsohn <dje.gcc@gmail.com> wrote:
>> The AIX system has not changed, so some change to GDB broke the build
>> on AIX.  It's warning about missing declarations. Is this related to
>> the C++ change?
>
> An aarch64 binutils patch from me broke it.  I can't see anything
> wrong with my patch.  I can't tell why AIX builds are broken without
> access to an AIX system, and my request for a compile farm account 2.5
> weeks ago hasn't gotten a response yet, so I'm stuck.
>
> https://sourceware.org/ml/binutils/2016-05/msg00103.html
> https://sourceware.org/ml/gdb-testers/2016-q2/msg03281.html
>
> Jim

Re: build bot failure

[Jim Wilson]

On Sun, Jun 12, 2016 at 10:27 AM, David Edelsohn <dje.gcc@gmail.com> wrote:
> The problem is the debug stabstring for bfd_reloc_code_real_names now
> overflows the maximum length.  Your patch overflowed the limit.
>
> More recent releases of GCC for AIX added support for stab
> continuation strings, but not the release installed on the machine
> running the buildbot.

I finally got my compile farm account over the weekend, and have
confirmed what David Edelsohn already said.  bimutils/gdb will build
without debug info, but with debug info the build fails because of a
bug in the gcc-4.8.1 release in /opt/freeware/bin.  This bug is fixed
in gcc-5.x.

I tried a build of gcc-5.x using the existing gcc-4.8.4 release, and
ran into problem with shared libraries.  For some reason, the
installed gcc insists on using libraries in the build tree instead of
libraries from the install tree, and then fails with a linker error.
This is apparently an AIX feature, and I don't know how to work around
it.  I also tried builds with xlc, and discovered that neither
binutils nor gcc builds with xlc out of the box.  Apparently no one is
testing this.

Since there is no problem with my patch, and since I don't have any
business or personal interest in AIX, I'm not planning to look into
this any further.

We could maybe request a new gcc install on the gcc111 AIX machine,
but I don't know if this will accomplish anything useful anytime soon.

Jim

Re: build bot failure

[Sergio Durigan Junior]

On Wednesday, June 22 2016, Jim Wilson wrote:

> On Sun, Jun 12, 2016 at 10:27 AM, David Edelsohn <dje.gcc@gmail.com> wrote:
>> The problem is the debug stabstring for bfd_reloc_code_real_names now
>> overflows the maximum length.  Your patch overflowed the limit.
>>
>> More recent releases of GCC for AIX added support for stab
>> continuation strings, but not the release installed on the machine
>> running the buildbot.
>
> I finally got my compile farm account over the weekend, and have
> confirmed what David Edelsohn already said.  bimutils/gdb will build
> without debug info, but with debug info the build fails because of a
> bug in the gcc-4.8.1 release in /opt/freeware/bin.  This bug is fixed
> in gcc-5.x.

Thanks for looking into this, Jim.

> Since there is no problem with my patch, and since I don't have any
> business or personal interest in AIX, I'm not planning to look into
> this any further.

Right.  This goes in the same direction as I am heading: reduce the
noise caused by some targets on our BuildBot.

Cheers,

-- 
Sergio
GPG key ID: 237A 54B1 0287 28BF 00EF  31F4 D0EB 7628 65FC 5E36
Please send encrypted e-mail if possible
http://sergiodj.net/

Re: build bot failure

[Yao Qi]

On Sun, Jun 12, 2016 at 6:27 PM, David Edelsohn <dje.gcc@gmail.com> wrote:
> Jim,
>
> The problem is the debug stabstring for bfd_reloc_code_real_names now
> overflows the maximum length.  Your patch overflowed the limit.
>
> More recent releases of GCC for AIX added support for stab
> continuation strings, but not the release installed on the machine
> running the buildbot.
>

Hi David,
I am still unable to build GDB on AIX (gcc111) due to this gcc/ld issue
today.  I know you've already started to address this issue (we talked
about this in Cauldron in July).  When can we get a new toolchain
installed on it or get it replaced by a new one?

I touch some rs6000 aix software single step code, and I'd like to test
GDB on aix to make sure my patches regression-free on aix.

-- 
Yao (齐尧)

Re: build bot failure

[David Edelsohn]

On Mon, Nov 14, 2016 at 6:05 AM, Yao Qi <qiyaoltc@gmail.com> wrote:
> On Sun, Jun 12, 2016 at 6:27 PM, David Edelsohn <dje.gcc@gmail.com> wrote:
>> Jim,
>>
>> The problem is the debug stabstring for bfd_reloc_code_real_names now
>> overflows the maximum length.  Your patch overflowed the limit.
>>
>> More recent releases of GCC for AIX added support for stab
>> continuation strings, but not the release installed on the machine
>> running the buildbot.
>>
>
> Hi David,
> I am still unable to build GDB on AIX (gcc111) due to this gcc/ld issue
> today.  I know you've already started to address this issue (we talked
> about this in Cauldron in July).  When can we get a new toolchain
> installed on it or get it replaced by a new one?
>
> I touch some rs6000 aix software single step code, and I'd like to test
> GDB on aix to make sure my patches regression-free on aix.

There is a new POWER8 AIX 7.2 system in the GNU Compile Farm.  That is
the system running the GDB Buildbot for AIX.

Patches can be tested in the "custom" builder of the Buildbot, or you
can log in to the system directly.

gcc119.fsffrance.org aka power8-aix.osuosl.org

Laurent and I have not formally announced it because of a question
about I/O tuning, but the system works and is available.

Thanks, David