865844 - test_WebGLContextLost from bug 844323 causes assertion failures

Status: NEW

(Core :: Graphics: Layers, defect)

Description

[Justin Lebar (not reading bugmail)]

See bug 844323 comment 45; this test (currently disabled) causes assertion failures.

It's not clear to me yet whether this issue needs to block the landing of bug 844323, but I kind of hope not, since that landing is blocking a lot of other, important B2G work.

Comment 1

[Nicolas Silva [:nical]]

It appears that in RenderFrameParent we hit the path where we can't get a LayerManagerComposite, here: https://mxr.mozilla.org/mozilla-central/source/layout/ipc/RenderFrameParent.cpp#609

with nsContentUtils::LayerManagerForDocument(doc) returning a BasicLayerManager here:
https://mxr.mozilla.org/mozilla-central/source/layout/ipc/RenderFrameParent.cpp#481

I don't know yet if we are supposed to get into this code path sometimes or if it always means we ran into an error.

Comment 2

[Nicolas Silva [:nical]]

we don't seem to ever hit the supposedly correct path 
https://mxr.mozilla.org/mozilla-central/source/layout/ipc/RenderFrameParent.cpp#607

where we have a LayerManagerComposite

Comment 3

[Nicolas Silva [:nical]]

Attachment: Make some global compositing information accessible without a compositor.

While looking at this bug it turned out that we were always falling back on the "oh darn, we don't have a LayerManagerComposite" code path in RenderFrameParent, so here is a patch that actually queries compositing info (layers backend and max texture size) without needing a layer manager composite.

This doesn't solve the problem, though.
Now it seems that we are in a different setup in this test than in the regular linux config: we do some IPC stuff and use PuppetWidget but we don't initialize OMTC. And then we still try to use the ClientLayerManager stuff but without a proper omtc setup it crashes in the Client code.

Comment 4

[Nicolas Silva [:nical]]

So, I don't know why test_WebGLContextLost uses a different setup than the Cipc/Ripc tests. My understanding is that it also means to run stuff in a b2g-like setup with IPC going on, except that Ripc/Cipc has "browser.tabs.remote" prefed on while test_WebGLContextLost has "dom.mozBrowserFramesEnabled" instead.

As a result, Cipc/Ripc initialize OMTC, while test_WebGLContextLost doesn't, even though it uses PuppetWidget that require OMTC. Hence the crash.

Comment 5

[Nicolas Silva [:nical]]

> As a result, Cipc/Ripc initialize OMTC, while test_WebGLContextLost doesn't,
> even though it uses PuppetWidget that require OMTC. Hence the crash.

Actually I take back what I said about OMTC init. It is more about the order of things happening. OMTC is setup in both cases.

Comment 6

[Nicolas Silva [:nical]]

Comment on attachment 744629 [details] [diff] [review]
Make some global compositing information accessible without a compositor.

Review of attachment 744629 [details] [diff] [review]:
-----------------------------------------------------------------

this is a bit outdated, and doesn't fix the problem entirely

Comment 7

[Milan Sreckovic [:milan] (needinfo for best results)]

Nical is swamped a bit, I'll find somebody else.

Comment 8

[Lorie Rost]

Attachment: same problem

Thanks from causlimilsa1979.