[Mesa-dev] Mesa 17.2.4 release candidate

[Ilia Mirkin]

On Mon, Oct 30, 2017 at 12:43 PM, Andres Gomez <agomez at igalia.com> wrote:
> On Mon, 2017-10-30 at 11:39 -0400, Ilia Mirkin wrote:
>> On Mon, Oct 30, 2017 at 11:29 AM, Emil Velikov <emil.l.velikov at gmail.com> wrote:
>> > On 28 October 2017 at 21:35, Andres Gomez <agomez at igalia.com> wrote:
>> > > On Fri, 2017-10-27 at 14:14 -0400, Ilia Mirkin wrote:
>> > > > On Fri, Oct 27, 2017 at 1:43 PM, Andres Gomez <agomez at igalia.com> wrote:
>> > > > > Rejected (6)
>> > > > > ============
>> > > > >
>> > > > > Ilia Mirkin (1):
>> > > > >       glsl: fix derived cs variables
>> > > > >
>> > > > > Reason: Commit is too big for stable at this point.
>> > > >
>> > > > The issue it fixes in regular compute shaders is slightly difficult to
>> > > > hit (but there are piglits that do now), however the issue it hits
>> > > > with ARB_compute_variable_group_size is fairly trivial to encounter.
>> > > >
>> > > > It seems silly to put out releases with known bugs when a fix is
>> > > > easily available and apply-able, with negligible risk of messing
>> > > > things up.
>> > > >
>> > > > Note that this all only affects nouveau and radeonsi, as those are the
>> > > > only drivers that make use of the lowering.
>> > >
>> > > Ilia, I understood by your answer [1] when I asked about your opinion
>> > > regarding 17.2.4 inclusion that it was OK to omit in this series but
>> > > that it should definitely be included in 17.3.
>> > >
>> > > Maybe I didn't make it clear that I was asking for the 17.2 queue and I
>> > > misunderstood your answer (?).
>> > >
>> > > Let's leave it as it is for this release and will see if we can include
>> > > it for the next one.
>> > >
>> >
>> > I'm inclined to agree with Andres - let's leave the patch out of 17.2.x.
>> >
>> > 17.3.x on the other hand is still fairly fresh, so I've pulled the
>> > patch for 17.3.0-rc2.
>>
>> In that case you should definitely remove
>> GL_ARB_compute_variable_group_size support from that release series.
>
> In that case, can we get?
>
>  * A specific less invasive solution for the 17.2 queue (don't know if
>    that's even possible).

Invasive is in the eye of the beholder. I don't think my change is
invasive at all. (But even if it were, that still shouldn't cause it
to not be backported into a stable release at developer request.)

>
> or
>
>  * A patch to disable the mentioned extension in the affected drivers.

Feel free to do that yourself, or find someone to do it for you. Doing
that will still have some bugs with regular compute shaders which
piglit will run into, although it's less clear that real applications
will hit it.

I think at this point, I will do what I originally said I'd do last
time this kind of issue came up -- stop caring about stable releases.
>From now on, I will no longer mark patches for stable (since doing so
only seems to cause me heartache), and make sure that anyone who's
trying to get help for mesa issues starts with git master, since the
stable releases aren't interested in all bug fixes. (And I certainly
don't have the time to re-debug issues that have already been fixed,
especially ones like this which I would never have figured out in a
full application, if it wasn't for the simple test case provided in
the referenced bug.)

Cheers,

  -ilia