make a shallow-copy so we don't get racy trying to lock/unlock our private bitmap

src/image/SkImage_Raster.cpp

Index: src/image/SkImage_Raster.cpp
diff --git a/src/image/SkImage_Raster.cpp b/src/image/SkImage_Raster.cpp
index af8f79010f134047b0ebf25678f9a22364dcb0d3..8afe7fd13aeec1668dd08c6ea15b1f985387c375 100644
--- a/src/image/SkImage_Raster.cpp
+++ b/src/image/SkImage_Raster.cpp
@@ -165,8 +165,11 @@ const void* SkImage_Raster::onPeekPixels(SkImageInfo* infoPtr, size_t* rowBytesP
 }
 
 void SkImage_Raster::onPreroll(GrContext* ctx) const {
-    fBitmap.lockPixels();
-    fBitmap.unlockPixels();
+    // SkImage can be called from lots of threads, but our fBitmap is *not* thread-safe,
+    // so we have to perform this lock/unlock in a non-racy way... we make a copy!
+    SkBitmap localShallowCopy(fBitmap);
+    localShallowCopy.lockPixels();
+    localShallowCopy.unlockPixels();

[f(malita), 2015/09/10 18:36:37]

This may solve the preRoll race, but don't we have the same issue when drawing
the same SkImage on multiple threads?  I presume internally we still call
lock/unlock on fBitmap in that case.

 }
 
 SkData* SkImage_Raster::onRefEncoded() const {