drm | Pipeline #4114 has failed for master | c55f1b9b

Lucas De Marchi lucas.de.marchi at gmail.com
Thu Sep 6 17:01:03 UTC 2018


On Thu, Sep 6, 2018 at 8:11 AM Emil Velikov <emil.l.velikov at gmail.com> wrote:
>
> HI all,
>
> On 6 September 2018 at 07:10, Lucas De Marchi <lucas.de.marchi at gmail.com> wrote:
> > On Wed, Sep 5, 2018 at 7:00 PM Rodrigo Vivi <rodrigo.vivi at gmail.com> wrote:
> >>
> >> well.... it builds for me.
> >>
> >> but any idea what might be wrong here on gitlab ci?
> >
> FWIW gitlab gives you the test target and command used. It might be a
> bit hard to find at first though :-\
>
> FAILED: meson-test
> /usr/bin/python3 /usr/local/bin/meson test --no-rebuild --print-errorlogs
>
>
> > From the logs:
> >
> > The output from the failed tests:
> >
> >  8/21 intel-symbol-check                      FAIL     0.29 s (exit status 1)
> >
> > --- command ---
> > NM='/usr/sbin/nm' /usr/sbin/bash
> > /builds/mesa/drm/_build/../intel/intel-symbol-check
> > intel/libdrm_intel.so.1.0.0
> > --- stdout ---
> > intel_get_genx intel_is_genx
> > -------
> >
> > So... it's not the build that fails, the test is checking for exported
> > symbols, and it only does that when using meson (actually as part of
> > the tests). And we don't use -fvisibility=hidden?!??! Ugh, I think we
> > need some fixing on that :(
> >
> I nuked it back in 0f8da82500ec542e269092c0718479e25eaff5f6 because
> weird compilers.
> I guess we can reconsider that?

I think we should reconsider. Having the default visibility as hidden
is the sanest thing to do to avoid problems of exporting what we
shoulnd't in a release. But even before that commit you pointed, it
only had the attributes to make a function private/public, but the
default was to export... do drm_public was basically a nop.

I know several libraries that use visibility=hidden for a very long
time and also maintain compatibility with several "embedded
toolchains". IMO If the compiler is that broken, then I think we'd
better workaround when building with that specific compiler if
possible or declare it as not supported.

>
> On the other hand: git grep does point out to drm_private. Will send a
> patch in a moment, unless someone beats me to it.

I missed that, thanks.  For a short-term fix, I think adding
drm_private attribute is reasonable, sorry for the breakage.

Lucas De Marchi

>
> -Emil



-- 
Lucas De Marchi


More information about the dri-devel mailing list