Prevent picture recording from over optimizing the culling of clips.

src/core/SkPictureRecord.cpp

Index: src/core/SkPictureRecord.cpp
diff --git a/src/core/SkPictureRecord.cpp b/src/core/SkPictureRecord.cpp
index 915736373d0aba5df1ac3116a09607984a006811..ec8012334ce8f5a59e687582692ca56dec7b12fd 100644
--- a/src/core/SkPictureRecord.cpp
+++ b/src/core/SkPictureRecord.cpp
@@ -479,6 +479,13 @@ static bool collapse_save_clip_restore(SkWriter32* writer, int32_t offset,
         return false;
     }
     SkASSERT(SAVE == op);

[robertphillips, 2013/08/13 12:12:32]

Add a kSaveSize near the kSaveLayerNoBoundsSize global and test that it is
correct in save()?

+    SkASSERT(8 == opSize);
+
+    // get the save flag (last 4-bytes of the space allocated for the opSize)
+    SkCanvas::SaveFlags saveFlag = (SkCanvas::SaveFlags) *writer->peek32(offset+4);

[robertphillips, 2013/08/13 12:12:32]

Is the evil combo we see everywhere a MatrixClip save? If not this could cause a
regression.

[djsollen, 2013/08/13 13:24:45]

Chrome doesn't do selective save and restores like this so yes we will still
catch all the evil cases.

+    if (SkCanvas::kMatrixClip_SaveFlag != (SkCanvas::kMatrixClip_SaveFlag & saveFlag)) {

[robertphillips, 2013/08/13 12:12:32]

// The following code is only correct for kMatrixClip style save's.
// TODO: set checkMatrix & checkClip booleans here and then check for the
offending operations in the following loop.

+        return false;
+    }
 
     // Walk forward until we get back to either a draw-verb (abort) or we hit
     // our restore (success).