589978 - [D3D9] 2D canvas layers always have premultiplied alpha

Status: RESOLVED FIXED

(Core :: Graphics, defect)

Comment 1

[Matt Woodrow (:mattwoodrow)]

Attachment: Fix

Fixes reftest image-element/canvas-outside-document.html

Comment 2

[Boris Zbarsky [:bzbarsky]]

Hmm.  Could we get away with not using premultiplied in some cases?  I thought we just did that because that's what cairo image surfaces do?

Comment 3

[Matt Woodrow (:mattwoodrow)]

2D content coming from canvas is a cairo surface. Don't know enough about this to say whether we could change the canvas backend to skip this.

Comment 4

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

This is correct; we should rename the flag to something like mDataIsPremultiplied.  The canvas 2d data is always premult, only WebGL content can be either.

Comment 5

[Bas Schouten (:bas.schouten)]

Comment on attachment 468524 [details] [diff] [review]
Fix

Let's do this, but let's do the rename Vlad suggested too.

Comment 6

[Matt Woodrow (:mattwoodrow)]

Attachment: Renamed var

Comment 7

[Bas Schouten (:bas.schouten)]

Comment on attachment 468584 [details] [diff] [review]
Renamed var

>diff --git a/gfx/layers/d3d9/ThebesLayerD3D9.cpp b/gfx/layers/d3d9/ThebesLayerD3D9.cpp
>--- a/gfx/layers/d3d9/ThebesLayerD3D9.cpp
>+++ b/gfx/layers/d3d9/ThebesLayerD3D9.cpp
>@@ -360,6 +360,17 @@ ThebesLayerD3D9::DrawRegion(const nsIntR
>   LayerManagerD3D9::CallbackInfo cbInfo = mD3DManager->GetCallbackInfo();
>   cbInfo.Callback(this, context, aRegion, nsIntRegion(), cbInfo.CallbackData);
> 
>+      nsRefPtr<gfxImageSurface> isurf =
>+      new gfxImageSurface(gfxIntSize(bounds.width, bounds.height),
>+                            gfxASurface::ImageFormatARGB32);
>+
>+      nsRefPtr<gfxContext> cx = new gfxContext(isurf);
>+
>+      cx->SetOperator(gfxContext::OPERATOR_SOURCE);
>+      cx->SetSource(destinationSurface);
>+      cx->Paint();
>+      void *bits = isurf->Data();
>+
>   if (UseOpaqueSurface(this)) {
>     surf->ReleaseDC(dc);
>   } else {

Looks like something spilled into the patch.

Comment 8

[Matt Woodrow (:mattwoodrow)]

Attachment: And without debugging code

Argh.

Comment 9

[[e-mail]]

Comment on attachment 468589 [details] [diff] [review]
And without debugging code

http://hg.mozilla.org/mozilla-central/rev/24ab75c424af