09:36 <kusma> zmike: I'm seeing a crash in a test here, where we get into zink_screen_timeline_wait() with a seemingly bogus batch_id (1513620848, but it seems to vary from run to run). This all seems to happen because we're trying to reclaim pb slabs that seems to have been free'd (they contain a refcount of 0 and mostly garbage)...

12:14 <zmike> 🤔

12:14 <zmike> seems bad

12:18 <zmike> pb bugs are the worst

13:16 <zmike> kusma: did you have further issues with https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15327

13:16 <kusma> zmike: No, looks good!

21:38 <ajax> zmike: https://gitlab.freedesktop.org/ajax/mesa/-/commits/kopper-directors-cut

21:38 <zmike> oh boy

21:39 <ajax> dropped the deadlock and renderdoc patches, split up the kopper bits

21:39 <ajax> interface, egl, glx, gallium core, zink

21:39 <ajax> i think the second-to-last doesn't actually produce a working zink driver but i'm not sure i care

21:39 <zmike> hm but I see you put kopper into swrast_loader_extensions

21:40 <zmike> which breaks the driver

21:40 <zmike> or have you fixed that in another way

21:40 <ajax> uh... if i did that was not intentional. possibly a git-rerere misfire.

21:41 <zmike> nvm I'm looking at the wrong thing

21:41 <zmike> ALL GOOD

21:41 <ajax> not sure if there's any further we can break down the last two that makes sense

21:41 <ajax> like i'm pretty sure the xcb workaround is a separate thing?

21:41 <zmike> xcb workaround?

21:42 <ajax> + <application name="Xwayland" executable="Xwayland">

21:42 <ajax> + <option name="disable_xcb_surface" value="true" />

21:42 <ajax> + </application>

21:42 <zmike> ah yeah

21:42 <zmike> that's just for running xwl

21:42 <zmike> so it could be split

21:43 <zmike> https://gitlab.freedesktop.org/ajax/mesa/-/commit/9d0e2126db3385d0e049f0c07282444433bdea4e#b4d0727a24f16f5eda402812ccf263a5022292d9_303_313 this part I think I had meant to key on disp->Options.Zink and never got around to it?

21:43 <zmike> I did it in x11 and not sure I got around to the others

21:44 <ajax> doesn't look like it's in wayland either

21:46 <zmike> ajax: actually while you're looking at this stuff I have this report https://gitlab.freedesktop.org/zmike/mesa/-/issues/109#note_1283599

21:46 <zmike> some kinda esoteric platform compile error?

21:46 <zmike> it's from kopper

21:48 <zmike> looks like you did some changes in the area, so maybe you fixed it?

21:49 <ajax> i had not fixed that yet

21:50 <ajax> i'm not sure how they get to that build path but it's an easy fix

21:50 <ajax> one moment

21:50 <zmike> nice

21:51 <zmike> https://gitlab.freedesktop.org/ajax/mesa/-/commit/5d72b49555863a94eddd20516e9fc0c87e70d0bb#6d12ed398ee79c3fa0d1e0968281c2ca378c4b09_53_57 <- I think all these can be removed since I put those into the base c_args a while back

21:53 <zmike> gitlab is really unhappy about showing me such huge diffs

21:54 <ajax> branch updated with all the above addressed

21:55 <zmike> cool

21:55 <ajax> and... gears crashes

21:55 <zmike> #executive

22:01 <ajax> ngh

22:14 <ajax> how much do we really care about keeping a working zink at every commit

22:15 <zmike> I don't really

22:15 <zmike> just mark it as "not zink bisectable" and move on

22:15 <zmike> I think anyone trying to bisect across this is going to readily notice if it breaks something anyway

22:17 <ajax> excellent

22:17 <ajax> because i'm trying to coax libglx into retrying zink as plain drisw if there's no kopper extension and i feel very unhappy about it

22:17 <ajax> but, if i don't, zink crashes instantly

22:17 <zmike> yeah let's not do that

22:18 <ajax> sweet

22:18 <ajax> then all that remains is to know why the last commit _also_ doesn't work

22:24 <ajax> https://paste.centos.org/view/raw/cb625ce9

22:25 <zmike> that's a puzzler

22:26 <ajax> oh dear

22:26 <ajax> https://paste.centos.org/view/raw/52c86c11

22:27 <ajax> odds that the deadlock patch was shielding me from whatever tf this is: non-zero?

22:27 <zmike> awkward

22:27 <zmike> uhhh

22:27 <zmike> which test is this?

22:27 <ajax> mesa devenv -C build glxgears

22:27 <zmike> deadlock patch afaik only fixed that one fbo blit piglit test

22:27 <zmike> and maybe like a couple other weird piglit cases

22:27 <zmike> nothing else

22:28 <zmike> I gotta run out for the night, but I can double check tomorrow

22:28 <zmike> (alternatively you can just test against the base copper branch)

22:28 <ajax> yeah, i'll poke a little more but i'm also out for the evening soon

22:29 <zmike> director's cut looks good overall though

22:30 <ajax> thanks. ready to have this landed

22:30 <ajax> eager, rather than ready, i suppose. you get me.

22:33 <zmike> amen