Counterproposal for skirting the BBH when the query fully contains the picture.

src/core/SkRecordDraw.cpp

Index: src/core/SkRecordDraw.cpp
diff --git a/src/core/SkRecordDraw.cpp b/src/core/SkRecordDraw.cpp
index bc5e7eb63feef19b231e4834f6c5a0ef3a8a1949..111e4d397794df7cb4e0c6b833da369c1914feff 100644
--- a/src/core/SkRecordDraw.cpp
+++ b/src/core/SkRecordDraw.cpp
@@ -18,8 +18,8 @@ void SkRecordDraw(const SkRecord& record,
         // Draw only ops that affect pixels in the canvas's current clip.
         SkIRect query;
 #if 1   // TODO: Why is this the right way to make the query?  I'd think it'd be the else branch.

[reed1, 2014/08/18 19:42:39]

Not sure I follow the comment.
Do we want to keep the #if 1 in the checked-in code?

[mtklein, 2014/08/18 19:45:24]

Yeah, I'd say so, at least until I understand why we build the BBH in
device-coordinates but seemingly make queries in user-space.  As it is, it
happens that GMs draw right this way and wrong in the else case, but I'd like to
leave a better explanation here.

[reed1, 2014/08/18 19:48:54]

Ah. Does the comment mean : "why isn't getClipDeviceBounds sufficient" ?

Lets say something about how to reproduce the problem here in the code if we're
going to have #else.

e.g.
/*
 * Seems like getClipDeviceBounds should be sufficient, but if we call it, some
GMs fail to draw correctly.
 * Lets figure out why : is that a bug in the GM, a bug in BBH, a bug in
getClipDeviceBounds, or other.
 */

-        SkRect clipBounds;
-        canvas->getClipBounds(&clipBounds);
+        SkRect clipBounds = { 0, 0, 0, 0 };
+        (void)canvas->getClipBounds(&clipBounds);
         clipBounds.roundOut(&query);
 #else
         canvas->getClipDeviceBounds(&query);