ICE failures with old GCC versions on the buildbots

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

ICE failures with old GCC versions on the buildbots

John W. Eaton
Administrator
I'm seeing internal compiler failures with GCC 4.9 and 5 on the buildbots:

GCC 4.9:

../../src/init2.c:52: MPFR assertion failed: p >= 2 && p <=
((mpfr_prec_t)((mpfr_uprec_t)(~(mpfr_uprec_t)0)>>1))
../src/libinterp/operators/op-scm-scm.cc: In function ‘MSparse<T>
quotient(const MSparse<T>&, const MSparse<T>&) [with T =
std::complex<double>]’:
../src/libinterp/operators/op-scm-scm.cc:253:1: internal compiler error:
Aborted
  }
  ^
Please submit a full bug report,
with preprocessed source if appropriate.
See <file:///usr/share/doc/gcc-4.9/README.Bugs> for instructions.
Preprocessed source stored into /tmp/ccPw5ALA.out file, please attach
this to your bugreport.

GCC 5:

../../src/init2.c:52: MPFR assertion failed: p >= 2 && p <=
((mpfr_prec_t)((mpfr_uprec_t)(~(mpfr_uprec_t)0)>>1))
In file included from ../src/liboctave/array/MSparse.h:123:0,
                  from ../src/liboctave/array/MatrixType.h:29,
                  from ../src/liboctave/array/dMatrix.h:31,
                  from ../src/liboctave/array/Range.h:30,
                  from ../src/libinterp/octave-value/ov.h:36,
                  from ../src/libinterp/octave-value/ovl.h:36,
                  from ../src/libinterp/operators/op-scm-scm.cc:29:
../src/liboctave/array/MSparse.cc: In function ‘MSparse<T>
quotient(const MSparse<T>&, const MSparse<T>&) [with T =
std::complex<double>]’:
../src/liboctave/array/MSparse.cc:478:1: internal compiler error: Aborted
  quotient (const MSparse<T>& a, const MSparse<T>& b)
  ^
Please submit a full bug report,
with preprocessed source if appropriate.
See <file:///usr/share/doc/gcc-5/README.Bugs> for instructions.


These failures seem to have started happening after I upgraded my
systems that are running the buildbot workers so it could be caused by
that and not a change in Octave.

It may be this bug:

   https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=888422

I can investigate and see whether there is a workaround, but I'm
wondering whether we should care about supporting builds with these
compilers.  GCC 5 was first released approximately 3 years ago.

jwe

Reply | Threaded
Open this post in threaded view
|

Re: ICE failures with old GCC versions on the buildbots

Mike Miller-4
On Fri, Mar 23, 2018 at 16:05:12 -0400, John W. Eaton wrote:
> I'm seeing internal compiler failures with GCC 4.9 and 5 on the buildbots:

I have my own daily build running on Ubuntu 14.04 with GCC 4.8 and I'm
not seeing any such errors.

> It may be this bug:
>
>   https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=888422

Maybe, are your buildbots Debian unstable with older compilers installed
on them?

If so, maybe a more appropriate way to test would be with containerized
distributions that were meant to work with GCC 4.8 or 4.9? A Debian 8
container for example?

> I can investigate and see whether there is a workaround, but I'm wondering
> whether we should care about supporting builds with these compilers.  GCC 5
> was first released approximately 3 years ago.

I don't mind dropping support for older compilers for Octave 5, but it
would be nice to at least ensure that Octave 4.4 works with GCC 4.8.

That may have nothing to do with whether you keep certain buildbot
configurations running.

--
mike

signature.asc (849 bytes) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: ICE failures with old GCC versions on the buildbots

John W. Eaton
Administrator
On 03/23/2018 04:20 PM, Mike Miller wrote:

> On Fri, Mar 23, 2018 at 16:05:12 -0400, John W. Eaton wrote:
>> I'm seeing internal compiler failures with GCC 4.9 and 5 on the buildbots:
>
> I have my own daily build running on Ubuntu 14.04 with GCC 4.8 and I'm
> not seeing any such errors.
>
>> It may be this bug:
>>
>>    https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=888422
>
> Maybe, are your buildbots Debian unstable with older compilers installed
> on them?

Yes, I'm using Debian testing with old versions of GCC still around.

> If so, maybe a more appropriate way to test would be with containerized
> distributions that were meant to work with GCC 4.8 or 4.9? A Debian 8
> container for example?

Yeah, that might be a better test anyway, building with old everything
instead of just an old gcc on an otherwise current system.

One way to do that would be a VM that I start and execute commands in
and then get the results back.  But if there is something simpler, I'm
up for that as well.

This also reminds me that I've been wanting to set up a VM running
Windows in a way that would allow running the test suite once Octave is
cross compiled for Windows.

> I don't mind dropping support for older compilers for Octave 5, but it
> would be nice to at least ensure that Octave 4.4 works with GCC 4.8.

OK.

jwe



Reply | Threaded
Open this post in threaded view
|

Re: ICE failures with old GCC versions on the buildbots

Mike Miller-4
On Fri, Mar 23, 2018 at 16:52:48 -0400, John W. Eaton wrote:
> One way to do that would be a VM that I start and execute commands in and
> then get the results back.  But if there is something simpler, I'm up for
> that as well.

The simplest would be to use a Debian chroot, and maybe run a separate
instance of buildbot with a different identifier in the chroot.

If you haven't done it recently, as root

    debootstrap jessie /path/to/somewhere/debian-jessie-chroot
    chroot /path/to/somewhere/debian-jessie-chroot
    apt-get update && apt-get install $things
    pip or whatever to set up buildbot

And you'll want to add a service or init script on the host to start the
additional chrooted buildbot.

--
mike

signature.asc (849 bytes) Download Attachment