[Intel-gfx] [RFC i-g-t v3 08/13] lib/stubs: Add stubs for intel_bufmgr.

[Emil Velikov]

On 22 June 2016 at 09:12, Daniel Vetter <daniel at ffwll.ch> wrote:
> On Tue, Jun 21, 2016 at 02:41:29PM +0100, Emil Velikov wrote:
>> On 21 June 2016 at 13:50, Marius Vlad <marius.c.vlad at intel.com> wrote:
>> > On Mon, Jun 20, 2016 at 03:52:35PM +0100, Emil Velikov wrote:
>> >> Hi Rob,
>> >>
>> >> A couple of nitpicks and a case of missing git add :-)
>> >>
>> >> On 15 June 2016 at 10:51,  <robert.foss at collabora.com> wrote:
>> >>
>> >> > +if HAVE_LIBDRM_INTEL
>> >> > +else
>> >> > +    libintel_tools_la_SOURCES +=       \
>> >> > +        stubs/drm/intel_bufmgr.c       \
>> >> > +        stubs/drm/intel_bufmgr.h
>> >> > +endif
>> >> > +
>> >> I believe I mentioned if before - please use the following construct
>> >>
>> >> if !HAVE_LIBDRM_INTEL
>> >>     libintel_tools_la_SOURCES +=       \
>> >>         stubs/drm/intel_bufmgr.c       \
>> >>         stubs/drm/intel_bufmgr.h
>> >> endif
>> >>
>> >> >  AM_CPPFLAGS = -I$(top_srcdir)
>> >> >  AM_CFLAGS = $(CWARNFLAGS) $(DRM_CFLAGS) $(PCIACCESS_CFLAGS) $(LIBUNWIND_CFLAGS) $(DEBUG_CFLAGS) \
>> >> >             -DIGT_SRCDIR=\""$(abs_top_srcdir)/tests"\" \
>> >> > diff --git a/lib/stubs/drm/README b/lib/stubs/drm/README
>> >> > new file mode 100644
>> >> > index 0000000..dec6a1d
>> >> > --- /dev/null
>> >> > +++ b/lib/stubs/drm/README
>> >> > @@ -0,0 +1,4 @@
>> >> > +intel_bufmgr.h is a copy the file provided in libdrm (intel/intel_bufmgr.h).
>> >> > +
>> >> s/copy the file provided in/local copy of the file provided by/
>> >>
>> >> Sadly that's not the case atm. There's a fair few changes in the local 'copy'.
>> >> Perhaps you forgot to git add ?
>> >>
>> >> > +Before releasing i-g-t a current copy of intel_bufmgr.h should be copied into
>> >> > +this directory of i-g-t.
>> >
>> > Auch, but synchronizing this wouldn't be a nightmare? I mean stubs
>> > might be added/removed in the future.
>> >
>> Maybe slightly annoying, but definitely not nightmare. As with any
>> other library one _cannot_ remove API without bumping major (and/or
>> changing the library name), so things should be fine on that regard.
>> Guess one should stress that intel_bufmgr.h should be copied from the
>> latest libdrm release and
>>
>> >>
>> >> Thinking about it... I'm not sure that the release manager will find
>> >> this note here :-\
>> >>
>> >> Daniel V, Marius, do you guys have a tick-off list somewhere where
>> >> this could be added ?
>> >
>> > README and NEWS. Don't know of any other.
>> >
>> Can I ask that you/Daniel can update those ? I'm afraid
>
> Instead of documenting, can't we enforce this with an autobuilder which
> runs with the stubbed-out libdrm always? Collabora has such a thing I
> heard ;-)
>
> Imo trying to fix up the build for releases is way too late, we should
> never have a broken build. Or at least not for long.
Ideally the git IGT users (with stubbed libdrm_intel) will promptly
provide feedback/patches ;-)

That aside - I fully agree that catching things earlier is the thing
do to. Props to Tomeu and others for setting up the CI/Jenkins
instance.

Thanks
Emil