<anholt>
err, driswCreateDisplay always succeeds, so you always go the zink route and don't fall back to non-zink (which is what I bet virpipe needs)?
<zmike>
is that a more recent job? last time I checked it was the virgl jobs failing
<zmike>
hmmm
<zmike>
could be?
<anholt>
I pulled up the mr you linked and looked at its contents and its last failing job
<zmike>
gotcha, I'm on mobile atm
<anholt>
so I think instead of that glx change you want driswCreateScreen to try zink even if !pdyp->zink, but fall back to swrast if that fails and !pdyp->zink
<anholt>
(pdpyp->zink begin "only accept zink and no substitutes"
<anholt>
*being
<zmike>
ah I think that might be an ajax tweak
<zmike>
will have to read and refresh
<zmike>
the issue with always trying zink is the sw will never be called
<zmike>
so for example just setting LIBGL_ALWAYS_SOFTWARE should use llvmpipe
<zmike>
but if I always try zink then it gets zink
<zmike>
this is sort of the one tricky case is was trying to handle
eukara has quit [Remote host closed the connection]
eukara has joined #zink
fahien has quit [Ping timeout: 480 seconds]
eukara has quit []
eukara has joined #zink
fahien has joined #zink
<ajax>
if something in ci/zink-radv-fails.txt xpasses, ci calls that a fail, right?
<zmike>
no, because that doesn't run on gitlab ci
<zmike>
only manual
<zmike>
unless that's changed?
<ajax>
lvp then
<zmike>
yep
<ajax>
then why do things in the fail list pass for me
<zmike>
could be flake?
<ajax>
hm
<anholt>
ajax: I've had experiences where gallivm results are different from checked in, presumably due differences in the cpu targeted.
fahien has quit [Ping timeout: 480 seconds]
<ajax>
i'm mostly looking at egl and glx failures, which their piglit tests tend not to do anything more complicated than flat shaded triangles, so i'd hope that kind of thing would be rare. but yeah.
<anholt>
if it's those, then I'd suspect your version of xvfb vs ci's.
<anholt>
we should really build newer X in CI, but ughhhh. i don't wanna.
<anholt>
(I wish we had a distro whose options included something between "irrelevantly old" and "some random maybe-fresh stuff every time you rebuild your container")
<anholt>
maybe ajax knows of some distro with a regular release cadence that likes to update the packages we care about? ;)
eukara has quit [Remote host closed the connection]
eukara has joined #zink
fahien has joined #zink
fahien has quit [Ping timeout: 480 seconds]
omegatron has quit [Quit: What happened? You quit!]