[PATCH xserver] configure.ac: unconditionally enable kdrive

[Eric Anholt]

Adam Jackson <ajax at redhat.com> writes:

> On Wed, 2017-11-01 at 21:50 -0700, Matt Turner wrote:
>> On Tue, Oct 24, 2017 at 10:27 AM, Matt Turner <mattst88 at gmail.com>
>> wrote:
>> > Reviewed-by: Matt Turner <mattst88 at gmail.com>
>> 
>> Please merge. :)
>
> Full disclosure: the reason I haven't is because it exposed that my
> test setup was failing to build Xephyr (for exactly the reason the
> patch fixes), and that glamor was now failing xts. I am quite sure it
> worked in 1.19 at the time, but building 1.19 [1] now was also failing
> xts, which points the finger at llvmpipe since it rules out any change
> in glamor as the culprit.
>
> I'd already worked around one such regression with llvmpipe [2] but
> this one has eluded me. Eric Anholt had written a script to run the
> test on a hardware-backed Xephyr, but didn't think we could integrate
> it with make check.
>
> I'd like to start doing that, though. In this order:
>
> 1) Add the script
> 2) Add it to make check, guarded by [ -z "$DISPLAY" ]
> 3) Apply this kdrive patch (assuming it passes on my hardware!)
>
> I think that preserves the greatest amount of bisectable history
> without introducing any known-broken tree states. It also means we'd be
> testing something more like what people actually run; the set of people
> using glamor backed by llvmpipe I suspect is zero, it's uniformly
> slower than fb.
>
> Sound reasonable?

My reasoning behind using llvmpipe only for make test was because it
meant that we'd all be running the same driver for make check.  My fear
was that if we use the HW drivers, it means that I check in some glamor
code that works on my driver, and keithp on his different gen of intel
(or somebody's radeon, or whatever) tries to run make check and the
build breaks because of some GL driver bug.

Now, it would be great if that meant that we were all making sure that
all of our personal GL drivers work for glamor.  Given that I haven't
even debugged the llvmpipe fails in all this time, I'm skeptical of that
working out, and instead we'd have people just ignoring the glamor tests
even more.

So, mixed feelings.  I think it's worth a try, and it might be good for
us in the end because the shorter test runtime on HW drivers makes it so
that we actually consistently run the tests and do the debug work.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 832 bytes
Desc: not available
URL: <https://lists.x.org/archives/xorg-devel/attachments/20171102/b0412ec6/attachment.sig>