Fix mixed sampled inverse path fills

src/gpu/batches/GrStencilAndCoverPathRenderer.cpp

Index: src/gpu/batches/GrStencilAndCoverPathRenderer.cpp
diff --git a/src/gpu/batches/GrStencilAndCoverPathRenderer.cpp b/src/gpu/batches/GrStencilAndCoverPathRenderer.cpp
index c96e2074c56136f2bbe04b16da979c16dcc10026..a11d2b468345120c3b499a5877aa812c016c2edd 100644
--- a/src/gpu/batches/GrStencilAndCoverPathRenderer.cpp
+++ b/src/gpu/batches/GrStencilAndCoverPathRenderer.cpp
@@ -135,6 +135,9 @@ bool GrStencilAndCoverPathRenderer::onDrawPath(const DrawPathArgs& args) {
             }
         }
         const SkMatrix& viewM = viewMatrix.hasPerspective() ? SkMatrix::I() : viewMatrix;
+        if (pipelineBuilder->getRenderTarget()->hasMixedSamples()) {
+            pipelineBuilder->disableState(GrPipelineBuilder::kHWAntialias_Flag);
+        }
         args.fTarget->drawNonAARect(*pipelineBuilder, args.fColor, viewM, bounds, invert);

[Chris Dalton, 2015/11/12 02:28:02]

Is "drawNonAARect" more of a perf hint, that says the DT doesn't need to worry
about AA? Or is it intended to be a guarantee that the draw will not be
antialiased?

[bsalomon, 2015/11/12 16:39:32]

It's supposed to be strict. It's often used to draw in device space where we
have a rectangle of pixels we want to cover, as in this use case. I'd like to
figure out how to remove or rework this method since it's definition isn't
orthogonal to the pipelinebuilder param it takes. Some of the callers shouldn't
really be calling at the draw target level, but should be using GrDrawContext
(e.g. clip mask manager). We've gone back and forth on whether it should be
possible for a batch to do this type of covering step so that things like this
code path wouldn't need two batches. It's a bit in flux and how that'd be
spec'ed isn't clear yet. It'd be nice if the path renderers could look more like
GrBatch factories where the batch handled the whole path render rather than
whacking on a grdrawtarget.

     } else {
         GR_STATIC_CONST_SAME_STENCIL(kStencilPass,