Revamp filter primitive region calculations for Filter Effects

third_party/WebKit/Source/platform/graphics/filters/FETile.cpp

Index: third_party/WebKit/Source/platform/graphics/filters/FETile.cpp
diff --git a/third_party/WebKit/Source/platform/graphics/filters/FETile.cpp b/third_party/WebKit/Source/platform/graphics/filters/FETile.cpp
index 8677ed991cbdee191da940562afbad4b0941fbc9..b777eb13ea2dd0d977600bf0d07bcf1977c76771 100644
--- a/third_party/WebKit/Source/platform/graphics/filters/FETile.cpp
+++ b/third_party/WebKit/Source/platform/graphics/filters/FETile.cpp
@@ -46,8 +46,12 @@ FloatRect FETile::mapPaintRect(const FloatRect& rect, bool forward) const
 sk_sp<SkImageFilter> FETile::createImageFilter()
 {
     sk_sp<SkImageFilter> input(SkiaImageFilterBuilder::build(inputEffect(0), operatingColorSpace()));
-    FloatRect srcRect = inputEffect(0)->filterPrimitiveSubregion();
-    FloatRect dstRect = applyEffectBoundaries(getFilter()->filterRegion());
+    FloatRect srcRect;
+    if (inputEffect(0)->getFilterEffectType() == FilterEffectTypeSourceInput)

[Stephen White, 2016/09/13 15:48:02]

Would it not be possible to have make filterPrimitiveSubregion() virtual, and
have SourceGraphic return filterRegion() as its filterPrimitiveSubregion()? It
would allow us to eliminate the special case here. Or is that a bad idea for
other reasons?

[fs, 2016/09/13 16:11:52]

I think that more general here would be to check clipsToBounds() (== has filter
primitive region), so specializing to SourceGraphic probably isn't necessarily
the right thing. (Also the filter [and hence region] of the source may differ
from the filter of the feTile in some cases.)

+        srcRect = getFilter()->filterRegion();
+    else
+        srcRect = inputEffect(0)->filterPrimitiveSubregion();
+    FloatRect dstRect = filterPrimitiveSubregion();
     return SkTileImageFilter::Make(srcRect, dstRect, std::move(input));
 }