1040917 - OMTC basic on GTK3 : Image surfaces make tcanvasmark's canvas3D test freeze with system-cairo

Status: RESOLVED WONTFIX

(Core :: Graphics: Layers, defect)

Description

[Frederic Plourde (:fred23)]

Only tested and reproduced so far with GTK3 builds with OMTC basic enabled (force-enable) and image offscreen surfaces enabled with "use-image-offscreen-surfaces" preference (see Bug 1015218)

steps to reproduce :
**********************
Apply abovementionned patches from Bug1015218
Turn on the abovementioned prefs
Build FF for GTK3 (system-cairo + system-pixman)

Locally run tcanvasmark talos test and you will notice that the test "freeze" at the "canvas3D" demo (where a series of 1-2-3-4 numbered boxes is rendered on a spiral path) after having drawn "some" boxes.

I don't know if the number of boxes correctly drawn before freeze is constant from one try to the other.

Comment 1

[Frederic Plourde (:fred23)]

Just as a precision, This bug was initially created as a means to explain some performance drops I was noticing on my local talos perf tests, but in the end, those performance drops turned out to be perf. *gains* (snarkfest was interpreting the results in the wrong way). 

So I don't know if this freeze I described above is even happening on the try servers.
That should be the first step, see if this is actually reproducible in the trys.

Comment 2

[Martin Stránský [:stransky] (ni? me)]

Reproducible with latest trunk, throws:

[GFX2-]: Invalid surface
[GFX2-]: DrawTargetCairo context in error state: out of memory(1)

Comment 3

[Martin Stránský [:stransky] (ni? me)]

I see the freeze on NVDIA graphics only. Intel drivers works without freeze.

Comment 4

[Martin Stránský [:stransky] (ni? me)]

It also causes flickering during scrolling on nvidia drivers. Intel seems to be fine.

Comment 5

[Martin Stránský [:stransky] (ni? me)]

I was wrong here. I run some tests on latest trunk during investigation of the OMTC crashes and it turns out that this freeze is produces the system-cairo library, as well as the OMTC crashes. With in-tree cairo, both (OMTC and offscreen surfaces) works well without crashes/hangs. The canvasmark test also passes with in-tree cairo no matter which driver is used (tested nvidia and intel).

Comment 6

[Martin Stránský [:stransky] (ni? me)]

I guess we can close this while it's caused by system-cairo and works fine with in-tree library, right?

Comment 7

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

We can leave it open, note that it's system-cairo, we have a handful of bugs like that.

Comment 8

[Darkspirit]

bug 1727876