i686 badly broken from premature 'python-build-system' merge

i686 badly broken from premature 'python-build-system' merge

[Mark H Weaver]

Hi,

yasm was apparently broken on i686 on the 'python-build-system' branch,
which was then subsequently merged into master.  As a result, many
important package are dependency-failures on i686, including all
gstreamer plugins, and thus all media players and web browsers.

This branch should not have been merged until these issues were fixed.
Now, i686 users won't be able to get important security updates such as
the recent icecat fixes (for which there are active exploits seen in the
wild) in a timely fashion.

     Mark

Re: i686 badly broken from premature 'python-build-system' merge

[Hartmut Goebel]

Am 01.12.2016 um 11:06 schrieb Mark H Weaver:
> yasm was apparently broken on i686 on the 'python-build-system' branch,

Can you please send me the ./test-suite.log. Given only the information
in the build log it is very hard to find the reason.

-- 
Regards
Hartmut Goebel

| Hartmut Goebel          | h.goebel@crazy-compilers.com               |
| www.crazy-compilers.com | compilers which you thought are impossible |

Re: i686 badly broken from premature 'python-build-system' merge

[Hartmut Goebel]

Am 01.12.2016 um 11:06 schrieb Mark H Weaver:
> yasm was apparently broken on i686 on the 'python-build-system' branch,

This error is not related to the python build system - until you proof
me wrong.

I've spend some time now for seeking the error (see below), until i
discovered that yasm is nor using the python build system at all. It is
not even using any python package beside "python-wrapper", which uses
the trivial build system and propagates "python".

Since I already did the analysis, here it is for the records:

I had a look at the source and I doubt this failure is related to the
python build system. I have no i686 system for testing this, though.

  *

    Any error related to the new python build system would occur when
    running any of the phases of the python build system. And even then
    only calls to "setup.py" are effected. It is very uncommon that this
    would happen in the middle of running the test suite.

  *

    The error occurs only for one series: "elf-amd64 objfmt", whereas
    all other series pass, e.g. "GAS elf-x86 objfmt". Both series call
    the same test-wrapper "out_test.sh" which seem not to have an
    case-handling for platforms. These two series only differ in the
    paths of the input and output files and the options passed to yasm.
    If this error would be related to the python build system, more if
    not all test-cases would fail.

  *

    The build log does not contain any entry related to python or
    setup.py after the patch-shebang phase. So the python related stuff
    seams not to be touched here.

-- 
Regards
Hartmut Goebel

| Hartmut Goebel          | h.goebel@crazy-compilers.com               |
| www.crazy-compilers.com | compilers which you thought are impossible |

Re: i686 badly broken from premature 'python-build-system' merge

[Ludovic Courtès]

Mark H Weaver <mhw@netris.org> skribis:

> yasm was apparently broken on i686 on the 'python-build-system' branch,
> which was then subsequently merged into master.  As a result, many
> important package are dependency-failures on i686, including all
> gstreamer plugins, and thus all media players and web browsers.

--8<---------------cut here---------------start------------->8---
$ ./pre-inst-env guix build yasm -s i686-linux -K
[…]
/gnu/store/7xv1zw247h3rlj8p6l95dz7frfypavcb-yasm-1.3.0
$ git describe
v0.11.0-3140-gd0b9c34
--8<---------------cut here---------------end--------------->8---

But now that I read Tobias message, I realize probably nasm is at fault,
not yasm.

Ludo’.

Re: i686 badly broken from premature 'python-build-system' merge

[Mark H Weaver]

Hi Ludovic,

ludo@gnu.org (Ludovic Courtès) writes:

> Mark H Weaver <mhw@netris.org> skribis:
>
>> yasm was apparently broken on i686 on the 'python-build-system' branch,
>> which was then subsequently merged into master.  As a result, many
>> important package are dependency-failures on i686, including all
>> gstreamer plugins, and thus all media players and web browsers.
>
> --8<---------------cut here---------------start------------->8---
> $ ./pre-inst-env guix build yasm -s i686-linux -K
> […]
> /gnu/store/7xv1zw247h3rlj8p6l95dz7frfypavcb-yasm-1.3.0
> $ git describe
> v0.11.0-3140-gd0b9c34
> --8<---------------cut here---------------end--------------->8---

Interesting.  Hydra tried to build it twice, and failed both times:

  https://hydra.gnu.org/build/1655354#tabs-buildsteps

I have saved logs from three build failures of 'yasm' on i686.  The
first one is from March 2016, and the other two are very recent.  There
are two tests that fail intermittently:

  FAIL: modules/objfmts/elf/tests/x32/elf_x32_test.sh
  FAIL: modules/objfmts/elf/tests/amd64/elf_amd64_test.sh

I also have saved logs from 'yasm' build failures on armhf and x86_64.
On armhf, the following test failed in July 2016:

  FAIL: modules/objfmts/elf/tests/gasx32/elf_gasx32_test.sh

and on x86_64, the following test failed in February 2016:

  FAIL: modules/objfmts/elf/tests/amd64/elf_amd64_test.sh

I restarted it again.  Maybe it will work on the third try.

> But now that I read Tobias message, I realize probably nasm is at fault,
> not yasm.

No.  Hydra indicates that 'yasm' is the reason for the important
dependency failures on i686:

  https://hydra.gnu.org/eval/109382?compare=109380#tabs-now-fail

Tobias may be right that 'nasm' is at fault for the non-Intel IceCat
jobs being removed, but that's apparently a different issue.

Note that there are also many new dependency failures on x86_64 due to a
build failure of 'gdm' on Hydra.  I guess that's yet another
non-deterministic build failure, since I managed to update my x86_64
system locally.

     Thanks,
       Mark

Re: i686 badly broken from premature 'python-build-system' merge

[Mark H Weaver]

I wrote:

> I restarted [the yasm build] again.  Maybe it will work on the third
> try.

Indeed, the it worked on the third try, and the 'gdm' build on x86_64
worked on the second try.  I'm restarting all dependency failures now.

     Mark

Re: i686 badly broken from premature 'python-build-system' merge

[Ludovic Courtès]

Mark H Weaver <mhw@netris.org> skribis:

> I wrote:
>
>> I restarted [the yasm build] again.  Maybe it will work on the third
>> try.
>
> Indeed, the it worked on the third try, and the 'gdm' build on x86_64
> worked on the second try.  I'm restarting all dependency failures now.

Weird.  Thanks for the update.

Ludo’.