Index: src/gpu/effects/GrXfermodeFragmentProcessor.cpp |
diff --git a/src/gpu/effects/GrXfermodeFragmentProcessor.cpp b/src/gpu/effects/GrXfermodeFragmentProcessor.cpp |
index c039db343315b6610ee1d4a62048becd2ffa2ba3..ab944d86fcbcb19f1bc157bf753fa135da503e76 100644 |
--- a/src/gpu/effects/GrXfermodeFragmentProcessor.cpp |
+++ b/src/gpu/effects/GrXfermodeFragmentProcessor.cpp |
@@ -72,28 +72,13 @@ private: |
GR_DEFINE_FRAGMENT_PROCESSOR_TEST(GrComposeTwoFragmentProcessor); |
const GrFragmentProcessor* GrComposeTwoFragmentProcessor::TestCreate(GrProcessorTestData* d) { |
-#if SK_ALLOW_STATIC_GLOBAL_INITIALIZERS |
// Create two random frag procs. |
- // For now, we'll prevent either children from being a shader with children to prevent the |
- // possibility of an arbitrarily large tree of procs. |
- SkAutoTUnref<const GrFragmentProcessor> fpA; |
- do { |
- fpA.reset(GrProcessorTestFactory<GrFragmentProcessor>::Create(d)); |
- SkASSERT(fpA); |
- } while (fpA->numChildProcessors() != 0); |
- SkAutoTUnref<const GrFragmentProcessor> fpB; |
- do { |
- fpB.reset(GrProcessorTestFactory<GrFragmentProcessor>::Create(d)); |
- SkASSERT(fpB); |
- } while (fpB->numChildProcessors() != 0); |
+ SkAutoTUnref<const GrFragmentProcessor> fpA(GrProcessorUnitTest::CreateChildFP(d)); |
+ SkAutoTUnref<const GrFragmentProcessor> fpB(GrProcessorUnitTest::CreateChildFP(d)); |
SkXfermode::Mode mode = static_cast<SkXfermode::Mode>( |
d->fRandom->nextRangeU(0, SkXfermode::kLastCoeffMode)); |
return SkNEW_ARGS(GrComposeTwoFragmentProcessor, (fpA, fpB, mode)); |
-#else |
- SkFAIL("Should not be called if !SK_ALLOW_STATIC_GLOBAL_INITIALIZERS"); |
- return nullptr; |
-#endif |
} |
GrGLFragmentProcessor* GrComposeTwoFragmentProcessor::onCreateGLInstance() const{ |