Many perl package failures on core-updates

Many perl package failures on core-updates

[Mark H Weaver]

On core updates, many perl packages fail with the error "Can't locate
Module/Build.pm in @INC".  I guess this is related to the fact that we
updated perl from 5.16.1 to 5.22.1, or perhaps due to the patches we've
applied to improve build determinism.  Can someone familiar with perl
take a look?

  http://hydra.gnu.org/build/937074
  http://hydra.gnu.org/build/935112
  http://hydra.gnu.org/build/942189
  http://hydra.gnu.org/build/943965
  http://hydra.gnu.org/build/936233
  http://hydra.gnu.org/build/935567
  http://hydra.gnu.org/build/937251
  http://hydra.gnu.org/build/934465
  http://hydra.gnu.org/build/944530
  http://hydra.gnu.org/build/934665
  http://hydra.gnu.org/build/935684
  http://hydra.gnu.org/build/936910
  http://hydra.gnu.org/build/934592

      Mark

Re: Many perl package failures on core-updates

[Ludovic Courtès]

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

> On core updates, many perl packages fail with the error "Can't locate
> Module/Build.pm in @INC".  I guess this is related to the fact that we
> updated perl from 5.16.1 to 5.22.1, or perhaps due to the patches we've
> applied to improve build determinism.  Can someone familiar with perl
> take a look?

I don’t really qualify as familiar with Perl ;-), but I’ve fixed some
(all?) of these in commit bb8afbf.

Thanks,
Ludo’.

Re: Many perl package failures on core-updates

[Andreas Enge]

Hello,

On Sun, Jan 17, 2016 at 09:44:37PM +0100, Ludovic Courtès wrote:
> I don’t really qualify as familiar with Perl ;-), but I’ve fixed some
> (all?) of these in commit bb8afbf.

there are still quite a few left, see
   http://hydra.gnu.org/eval/108793?full=1&compare=master#tabs-now-fail

Also, there are masses of failures for ghc packages and some for python
packages.

I am looking at the failure of doxygen; it should be resolved by switching
to a newer version that is compatible with flex-2.6.

Another show-stopper is qt-5, with an error message related to webkit.

Andreas

Re: Many perl package failures on core-updates

[Eric Bavier]

On Tue, 19 Jan 2016 19:41:12 +0100
Andreas Enge <andreas@enge.fr> wrote:

> Hello,
> 
> On Sun, Jan 17, 2016 at 09:44:37PM +0100, Ludovic Courtès wrote:
> > I don’t really qualify as familiar with Perl ;-), but I’ve fixed some
> > (all?) of these in commit bb8afbf.
> 
> there are still quite a few left, see
>    http://hydra.gnu.org/eval/108793?full=1&compare=master#tabs-now-fail

I'm working through these.

`~Eric

Re: Many perl package failures on core-updates

[Andreas Enge]

On Tue, Jan 19, 2016 at 07:41:12PM +0100, Andreas Enge wrote:
> I am looking at the failure of doxygen; it should be resolved by switching
> to a newer version that is compatible with flex-2.6.

I pushed an updated package, which changed more than expected, since now only
the cmake build system is supported.

I also dropped the propagated input graphviz; if this poses a problem
somewhere, we will need to revert this commit. In any case, graphviz is
not among the references of the doxygen package and not needed during
its build.

Andreas

Re: Many perl package failures on core-updates

[Ludovic Courtès]

Andreas Enge <andreas@enge.fr> skribis:

> I also dropped the propagated input graphviz; if this poses a problem
> somewhere, we will need to revert this commit. In any case, graphviz is
> not among the references of the doxygen package and not needed during
> its build.

I guess Graphviz was propagated so that ‘doxygen’ can produce class
diagrams out-of-the-box (it probably looks for ‘dot’ in $PATH.)  So I
would probably keep it propagated.

Ludo’.

Re: Many perl package failures on core-updates

[Andreas Enge]

On Wed, Jan 20, 2016 at 11:15:04PM +0100, Ludovic Courtès wrote:
> I guess Graphviz was propagated so that ‘doxygen’ can produce class
> diagrams out-of-the-box (it probably looks for ‘dot’ in $PATH.)  So I
> would probably keep it propagated.

I do not remember anymore, although it was me to do it (well, in April 2014).
At some point in time doxygen depended on graphviz, then the dependency
was dropped according to the changelog on the web site. But this was before
I added it, so you may be right. Does it have any impact on packages
that depend on doxygen?

I wonder if this is not a case where we should leave it to the user to
install or not the additional program, since doxygen should still be
functional without this feature?

Andreas

Re: Many perl package failures on core-updates

[Ludovic Courtès]

Andreas Enge <andreas@enge.fr> skribis:

> On Wed, Jan 20, 2016 at 11:15:04PM +0100, Ludovic Courtès wrote:
>> I guess Graphviz was propagated so that ‘doxygen’ can produce class
>> diagrams out-of-the-box (it probably looks for ‘dot’ in $PATH.)  So I
>> would probably keep it propagated.
>
> I do not remember anymore, although it was me to do it (well, in April 2014).
> At some point in time doxygen depended on graphviz, then the dependency
> was dropped according to the changelog on the web site. But this was before
> I added it, so you may be right. Does it have any impact on packages
> that depend on doxygen?

Dunno.

> I wonder if this is not a case where we should leave it to the user to
> install or not the additional program, since doxygen should still be
> functional without this feature?

Could be.  OTOH, there’s a risk of Doxygen being effectively “broken by
default” if, as I presume, many programs expect it to be able to
generate PNGs for class diagrams.

I’m chatting, but again, I haven’t actually tried.  :-)

Ludo’.

Re: Many perl package failures on core-updates

[Andreas Enge]

On Thu, Jan 21, 2016 at 09:48:58PM +0100, Ludovic Courtès wrote:
> Could be.  OTOH, there’s a risk of Doxygen being effectively “broken by
> default” if, as I presume, many programs expect it to be able to
> generate PNGs for class diagrams.

Okay. As propagated inputs are a somewhat dirty affair, I suggest to do
nothing until someone complains; and then we can add graphviz again with
a comment why this is needed.

Andreas

Re: Many perl package failures on core-updates

[Ludovic Courtès]

Andreas Enge <andreas@enge.fr> skribis:

> On Thu, Jan 21, 2016 at 09:48:58PM +0100, Ludovic Courtès wrote:
>> Could be.  OTOH, there’s a risk of Doxygen being effectively “broken by
>> default” if, as I presume, many programs expect it to be able to
>> generate PNGs for class diagrams.
>
> Okay. As propagated inputs are a somewhat dirty affair, I suggest to do
> nothing until someone complains; and then we can add graphviz again with
> a comment why this is needed.

Fine with me, we’ll see how it goes.

Ludo’.