add getTextureHandle to SkSurface

src/image/SkSurface_Gpu.cpp

Index: src/image/SkSurface_Gpu.cpp
diff --git a/src/image/SkSurface_Gpu.cpp b/src/image/SkSurface_Gpu.cpp
index fa006a61cdb4f7acf421a7fae3467ae70c063e80..c05fddafd0ebc562521dea85d7b37a60ad2d7b22 100644
--- a/src/image/SkSurface_Gpu.cpp
+++ b/src/image/SkSurface_Gpu.cpp
@@ -26,6 +26,19 @@ SkSurface_Gpu::~SkSurface_Gpu() {
     fDevice->unref();
 }
 
+GrBackendObject SkSurface_Gpu::onGetTextureHandle(ContentChangeMode mode) {

[Justin Novosad, 2015/06/26 21:33:15]

Where in this method does the genID get bumped? API should allow the the caller
to specify whether the texture is accessed for writing to avoid unnecessarily
bumping the genID.

[reed1, 2015/06/27 17:56:12]

New enum supports read and write access.

We are not bumping a genID, since there is no "bitmap" behind a gpu-surface (or
there shouldn't be). The way to access a surface's contents is via its snapshot,
and that will be invalidated.

+    GrRenderTarget* rt = fDevice->accessRenderTarget();
+    switch (mode) {
+        case kDiscard_ContentChangeMode:
+            rt->discard();

[Justin Novosad, 2015/06/26 21:33:15]

What is the use case for this?

[reed1, 2015/06/27 17:56:12]

Enum has changed, but the idea is to have 2 write access modes, with the option
for the caller to tell us they plan to overwrite the entire backing texture, in
which case we can discard rather than flush any pending writes.

+            break;
+        case kRetain_ContentChangeMode:
+            rt->prepareForExternalRead();   // todo: rename to prepareForExternalAccess()
+            break;
+    }
+    return rt->asTexture()->getTextureHandle();
+}
+
 SkCanvas* SkSurface_Gpu::onNewCanvas() {
     SkCanvas::InitFlags flags = SkCanvas::kDefault_InitFlags;
     // When we think this works...