detect fixed overflow in drawPoints

This particular test this is addressing is:

1. very large device (>32K in a test case)
2. drawPoints has special (fast) case where it converts all of the pts (rects) into fixed point
3. overflows fixed -- assert

Paths already don't draw into a device this large (they are limited by 32K I think, smaller if aa)
The (theoretical) solution for paths is to tile the draw itself. Perhaps that approach could be
applied everywhere in bitmapdevice...

This older issue sums up this idea: https://bugs.chromium.org/p/skia/issues/detail?id=2122
Will look into this general solution in a separate CL/effort.

Bug: skia:7425
Change-Id: I57f8da92df78127e6b8e42e422c43e50bb5748d6
Reviewed-on: https://skia-review.googlesource.com/96700
Commit-Queue: Mike Reed <reed@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>

1 files changed, 7 insertions, 0 deletions

diff --git a/src/core/SkRectPriv.h b/src/core/SkRectPriv.h
index 16c9535cc8..d2a2ce4665 100644
--- a/src/core/SkRectPriv.h
+++ b/src/core/SkRectPriv.h
@@ -9,6 +9,7 @@
 #define SkRectPriv_DEFINED
 
 #include "SkRect.h"
+#include "SkMathPriv.h"
 
 class SkRectPriv {
 public:
@@ -45,6 +46,12 @@ public:
         r->fTop    = SkMinScalar(pt.fY, r->fTop);
         r->fBottom = SkMaxScalar(pt.fY, r->fBottom);
     }
+
+    // conservative check. will return false for very large values that "could" fit
+    static bool FitsInFixed(const SkRect& r) {
+        return SkFitsInFixed(r.fLeft) && SkFitsInFixed(r.fTop) &&
+               SkFitsInFixed(r.fRight) && SkFitsInFixed(r.fBottom);
+    }
 };