[Webkit-unassigned] [Bug 63871] [Chromium] canvas/philip/tests/2d.gradient.object.update.html is failed on chromium.

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=63871





--- Comment #7 from Stephen White <senorblanco at chromium.org>  2011-07-08 10:43:22 PST ---
(In reply to comment #6)
> I agree this looks very invasive. Looking further up, I wonder why skia has this dual between graphicscontext->m_state (which has the gradient) and platformcontext which has the skia gradient. I notice the the webkit Gradient object itself holds a cache of the skia gradient, and it is correctly purged when addColorStop is called. If skia always went directly to m_state.fillGradient, I think we could skip all of these pushes.
> 
> Another clue that something funny is going on. Look in GraphicsContext.cpp
> 
> 
> #if !USE(SKIA)
> void GraphicsContext::setPlatformFillGradient(Gradient*)
> {
> }
> 
> void GraphicsContext::setPlatformFillPattern(Pattern*)
> {
> }
> 
> void GraphicsContext::setPlatformStrokeGradient(Gradient*)
> {
> }
> 
> void GraphicsContext::setPlatformStrokePattern(Pattern*)
> {
> }
> #endif
> 
> All of these seem to exist solely for skia. If we didn't make a (premature) copy of the gradient, but reached directly into m_state each time, I think we would (a) fix this bug, and (b) avoid this whole layer of indirection/copying.

I'm to blame for introducing these, in:

http://trac.webkit.org/changeset/46017

It was to refactor a bunch of cut-and-paste code, so that setupPaintForStroking() and setupPaintForFilling() could handle patterns and gradients, rather than having to cut-and-paste the same 5 lines of code we were using everywhere else for gradients and patterns (see the CL desc).

That may not be relevant anymore; I'm not sure.  But anything we come up with should be checked against the tests that were fixed in that CL, at least.

-- 
Configure bugmail: https://bugs.webkit.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.