<ajax>
since you get the ust back from the server in the notify event, you could estimate how long round trips are taking and do some exponential backoff to keep cpu overhead down
<ajax>
or you could just sleep for 1ms and say good enough. which has some appeal, i'm bored with this problem.
<zmike>
seems like it just might work
<ajax>
anyone who wanted to test it please do, i'd like at least some mitigation here
<zmike>
building
fahien has quit [Quit: fahien]
neobrain has left #zink [#zink]
<zmike>
ajax: I don't recall - were you actively looking at the glx visual fails from ci?
* zmike
has too much split brain this week
Akari has quit [Read error: Connection reset by peer]
Akari has joined #zink
<ajax>
zmike: stencil i thought we'd sorted. depth i still hadn't dug into.
<zmike>
#zink-ci is telling me that neither is sorted
<ajax>
are we still using non-threadsafe libX11 in ci because
<zmike>
oh true
<zmike>
I think so
<zmike>
iirc we're running debian sarge there still
<ajax>
i'm seriously tempted to just jam an LD_PRELOAD into the zink ci that does XInitThreads() from the constructor
<zmike>
hm I'm not seeing a deadlock in piglit/egl-nok-texture-from-pixmap at all?
<zmike>
so I don't know whether your patch affects it
<anholt>
I saw a job fail in glx-visuals-stencil recently.
<zmike>
I'm seeing them both fail very regularly
<ajax>
the egl-nok-tfp test was giving me the hang-at-exit in DestroySwapchain pretty reliably
<zmike>
odd
<ajax>
for whatever reason, this machine is very good at provoking races nobody else hits
<ajax>
i am not complaining, races you can reproduce are races you can fix. but it's definitely a bit eerie.