allow imagefilter to manage CTM decomposition

src/core/SkImageFilter.cpp

Index: src/core/SkImageFilter.cpp
diff --git a/src/core/SkImageFilter.cpp b/src/core/SkImageFilter.cpp
index 4128ba6cc20199f4489debfeb11a09add85609bc..6c73be29b59ff7bb0c5f7550b874854474c34853 100644
--- a/src/core/SkImageFilter.cpp
+++ b/src/core/SkImageFilter.cpp
@@ -330,6 +330,24 @@ bool SkImageFilter::asAColorFilter(SkColorFilter** filterPtr) const {
     return true;
 }
 
+bool SkImageFilter::onCanHandleAffine() const {

[Stephen White, 2016/04/25 14:09:34]

Wouldn't it be simpler to have (non-virtual) canHandleAffine do the recursion,
and make onCanHandleAffine non-recursive? That would keep the subclasses
simpler. Is there a node which needs to control the recursion beyond "I can't
handle affine transforms; abort"?

[reed1, 2016/04/25 14:13:35]

Good idea! See follow-on CL https://codereview.chromium.org/1915943002/

+    bool hasInputs = false;
+
+    const int count = this->countInputs();
+    for (int i = 0; i < count; ++i) {
+        SkImageFilter* input = this->getInput(i);
+        if (input) {
+            if (!input->canHandleAffine()) {
+                return false;
+            }
+            hasInputs = true;
+        }
+    }
+    // We return true iff we had 1 or more inputs, and all of them can handle affine.
+    // If we have no inputs, or 1 or more of them do not handle affine, then we return false.

[Stephen White, 2016/04/25 14:09:34]

Why would a null input imply we can't handle affine transforms? This just means
we're using the input primitive, which is pretty common.

+    return hasInputs;
+}
+
 bool SkImageFilter::applyCropRect(const Context& ctx, const SkIRect& srcBounds,
                                   SkIRect* dstBounds) const {
     SkIRect temp = this->onFilterNodeBounds(srcBounds, ctx.ctm(), kForward_MapDirection);