17:44 <zmike> ajax: ping

17:48 <zmike> I guess I'll create a ticket to discuss this awfulness

18:36 <ajax> zmike: hey

18:36 <zmike> oh there you are

18:37 <ajax> sorry, working weird hours these days

18:37 <zmike> I get that

18:39 <zmike> been back in the kopper mines this week

18:39 <zmike> would not recommend

18:52 <zmike> ajax: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17147 probably good to go now, though I'm wondering if I have a similar issue somewhere in egl

18:56 <ajax> i don't totally follow the explanation in #6713

18:56 <zmike> the swapchain resource is getting destroyed

18:56 <ajax> why would un-make-current-ing a framebuffer destroy its buffers? they live as long as the glxdrawable lives.

18:57 <zmike> right

18:57 <zmike> but the drawable is getting destroyed

18:58 <ajax> uh... at the end of the test, sure, there'a glXDestroyWindow there. is there another one happening implicitly?

18:58 <zmike> every time it binds a new context it destroys the old drawables

18:59 <zmike> it calls through to driReleaseDrawables and then the drawable gets destroyed

18:59 <ajax> wrong to do so, i think. both the context and the glxdrawable logically take a reference on the underlying framebuffer.

19:00 <ajax> should, anyway. if they're not then start there i think

19:00 <zmike> they do

19:00 <zmike> the glxdrawable is destroyed though

19:00 <zmike> so bye-bye

19:01 <ajax> you're saying the internal state backing the glxdraw gets blown away? not that the app actually calls glXDestroyWindow when you hit this.

19:01 <zmike> yep

19:01 <ajax> ngh

19:02 <zmike> if you check out zmike/zink-kopper you'll see it in action

19:06 <ajax> this all got way uglier while i wasn't looking

19:06 <ajax> what is this zombie drawable nonsense

19:07 <ajax> (i'm well aware that i probably approved the patch that added it)

19:07 <zmike> it's the only way

19:14 <ajax> so it's kind of insanely awful that the only thing that creates that state is driFetchDrawable

19:14 <zmike> yep

19:15 <ajax> well. that the only place we call that is from MakeCurrent. we ought to just do that from Create/DestroyDrawable too

19:15 <ajax> because Fetch does bump the refcount like we want...

19:34 <ajax> zmike: congrats on nerdsniping me

19:35 <ajax> holy cow is there a lot of broken shit in here

19:35 <zmike> yeah it's awful

19:35 <zmike> I spent like 5 hours on it this morning

19:38 <ajax> is there any deadline pressure on this?

19:38 <zmike> like it has to be done tomorrow?

19:38 <zmike> no

19:39 <ajax> i have a few other branches with work in the area here

19:39 <ajax> feeling the pull between fix it right and fix it soon

19:39 <zmike> I'm not in a rush

19:40 <zmike> this test fails now anyway

19:48 <zmike> ajax: in the meanwhile https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17147