Make ClipStack honor save flags

src/core/SkCanvas.cpp

Index: src/core/SkCanvas.cpp
===================================================================
--- src/core/SkCanvas.cpp	(revision 10709)
+++ src/core/SkCanvas.cpp	(working copy)
@@ -720,8 +720,9 @@
     newTop->fNext = fMCRec;
     fMCRec = newTop;
 
-    fClipStack.save();
-    SkASSERT(fClipStack.getSaveCount() == this->getSaveCount() - 1);
+    if (SkCanvas::kClip_SaveFlag & flags) {
+        fClipStack.save();
+    }
 
     return saveCount;
 }
@@ -896,7 +897,8 @@
     fDeviceCMDirty = true;
     fLocalBoundsCompareTypeDirty = true;
 
-    fClipStack.restore();
+    SkRasterClip* oldRasterClip = fMCRec->fRasterClip;
+
     // reserve our layer (if any)
     DeviceCM* layer = fMCRec->fLayer;   // may be null
     // now detach it from fMCRec so we can pop(). Gets freed after its drawn
@@ -907,6 +909,10 @@
     fMCStack.pop_back();
     fMCRec = (MCRec*)fMCStack.back();
 
+    if (NULL == fMCRec || fMCRec->fRasterClip != oldRasterClip) {

[bsalomon, 2013/08/14 15:49:13]

eek.. don't we want to stop computing raster clip for the GPU?

Is the problem than the MCRec doesn't remember the flags? Can it?

[robertphillips, 2013/08/14 16:16:58]

That is the problem.

[reed1, 2013/08/14 16:51:31]

If we pass the flags (or a subset of them) to clipstack.save(), would that make
the restore easier? (at least for the caller)

[robertphillips, 2013/08/14 17:00:16]

So a ClipStack frame would store the number of restores to ignore before really
restoring itself?

[bsalomon, 2013/08/14 18:23:28]

yeah I don't see why clip stack should see saves that don't affect it. It seems
legit to have the caller (SkCanvas) examine the flags to decide whether to
save/restore the clipstack.

+        fClipStack.restore();
+    }
+
     /*  Time to draw the layer's offscreen. We can't call the public drawSprite,
         since if we're being recorded, we don't want to record this (the
         recorder will have already recorded the restore).