1319303 - Fix GLXBadDrawable abort during running reftest on linux

Status: RESOLVED FIXED

(Core :: Graphics: WebRender, defect)

Description

[Sotaro Ikeda [:sotaro]]

When I run restest with "./mach restet", it caused GLXBadDrawable abort.

Comment 1

[Sotaro Ikeda [:sotaro]]

The bug is similar to Bug 1263515. X_GLXMakeCurrent was called after widget destruction.

Comment 2

[Sotaro Ikeda [:sotaro]]

Attachment: crash_stack

Comment 3

[Sotaro Ikeda [:sotaro]]

Attachment: patch - Add WebRenderBridgeParent::ClearResources()

By clean up GLContext before Widget destruction, the abort was addressed.

Comment 4

[Kartikaya Gupta (email:[e-mail])]

Next time please make sure the patch has a commit message and author information. I could have landed this for you.

Comment 5

[Sotaro Ikeda [:sotaro]]

(In reply to Kartikaya Gupta away[Nov24,Dec5) (email:[e-mail]) from comment #4)
> Next time please make sure the patch has a commit message and author
> information. I could have landed this for you.

Thanks. I am going to add them next time.

Comment 6

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/projects/graphics/rev/39388a579208
Add WebRenderBridgeParent::ClearResources() r=kats

Comment 7

[Kartikaya Gupta (email:[e-mail])]

So actually it's not clear to me now why this is supposed to fix the problem. ActorDestroy should be getting called *after* RecvDestroy, so if RecvDestroy is "too late" (i.e. after widget destruction) then cleaning up in ActorDestroy won't do any good because it will be even later.

In my latest try push I'm still seeing a crash with this same stack [1].

[1] https://treeherder.mozilla.org/logviewer.html#?job_id=31677637&repo=try#L1612

Comment 8

[Kartikaya Gupta (email:[e-mail])]

I think it might just be that WebRenderLayerManager is calling WRBridge()->SendDestroy() too late. I'll file a new bug for it.

Comment 9

[Sotaro Ikeda [:sotaro]]

(In reply to Kartikaya Gupta (email:[e-mail]) from comment #7)
> So actually it's not clear to me now why this is supposed to fix the
> problem. ActorDestroy should be getting called *after* RecvDestroy, so if
> RecvDestroy is "too late" (i.e. after widget destruction) then cleaning up
> in ActorDestroy won't do any good because it will be even later.
> 
> In my latest try push I'm still seeing a crash with this same stack [1].
> 

Bug 1320277 handled delayed message problem.