11:44 <kusma> zmike: The surface cache is holding pointers to freed memory...

11:45 <kusma> https://gitlab.freedesktop.org/-/snippets/2407

12:04 <kusma> zmike: I found this while debugging the spurious crash I get on the arb_shader_image_load_store-invalid Piglit with Zink + Lavapipe in !10467

12:11 <kusma> I created a ticket about it: https://gitlab.freedesktop.org/mesa/mesa/-/issues/5050

13:08 <zmike> I'm on it

14:17 <zmike> kusma: I think you prob want to mark that test as skip in ci for now to avoid adding flakes?

14:18 <kusma> zmike: Yeah, that's a fair point.

15:28 <zmike> this bug is way more frustrating than I expected

15:29 <zmike> it doesn't happen at all on zink-wip, but it's also not the bug I thought it was

15:29 <zmike> and it's infrequent enough that bisecting is useless

15:29 <kusma> Yeah, I had to run like 10 times to trigger it

15:30 <zmike> sometimes it takes more like 20-30

15:30 <kusma> Uggh.

15:30 <zmike> I think I'm closing in on something at last

15:30 <kusma> Great :)

15:30 <zmike> but definitely not a favorite bug so far

15:31 <zmike> aha

15:31 <zmike> I got it

15:33 <zmike> I don't understand how I haven't seen this before though

16:56 <zmike> kusma: okay, I got it, and it was way, way, way, way dumber than I thought

22:31 <zmike> \o/

23:06 <jekstrand> You mean you managed to get an MR in without AMD falling over?

23:12 <zmike> the key is you have to sit there watching it

23:12 <zmike> and the second you think the amd job failed

23:12 <zmike> you retry a lavapipe job

23:12 <zmike> and then retry the amd job

23:13 <jekstrand> *sigh*

23:13 <zmike> it's like blowing on the dice before you throw them

23:14 <jekstrand> ah

23:27 <omegatron> we all love determinism, do we?

23:28 <zmike> kusma: I think you might want to get your article updated to stop telling people that jekstrand supports advanced blend https://www.collabora.com/news-and-blog/blog/2021/06/14/zink-summer-2021-update/