cc: Added raster source.

cc/resources/raster_source.h

+// Copyright 2014 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#ifndef CC_RESOURCES_RASTER_SOURCE_H_
+#define CC_RESOURCES_RASTER_SOURCE_H_
+
+#include <vector>
+
+#include "base/memory/ref_counted.h"
+#include "cc/base/cc_export.h"
+#include "cc/debug/rendering_stats_instrumentation.h"

[reveman, 2014/10/27 14:39:46]

Can you forward declare RenderingStatsInstrumentation instead?

[vmpstr, 2014/10/27 15:55:11]

Done.

+#include "third_party/skia/include/core/SkCanvas.h"

[reveman, 2014/10/27 14:39:46]

forward declare SkCanvas instead?

[vmpstr, 2014/10/27 15:55:11]

Done.

+
+namespace cc {
+
+class CC_EXPORT RasterSource : public base::RefCountedThreadSafe<RasterSource> {
+ public:
+  struct CC_EXPORT Analysis {
+    Analysis() : is_solid_color(false) {}
+    ~Analysis() {}
+
+    bool is_solid_color;
+    SkColor solid_color;
+  };

[reveman, 2014/10/27 14:39:46]

Seems a bit silly to keep this struct when it's only used for solid color
detection. Why don't we rename AnalyzeInRect to IsSolidColor() instead?

[vmpstr, 2014/10/27 15:55:10]

I've kept this for now and added a TODO. Currently, the nice thing about the
struct is that we transport one opaque-ish variable instead of a bool and a
color. Secondly, if we do want to extend this, this currently requires no extra
plumbing, as all relevant information can be added to analysis. If we do decide
that we'll never really have any other type of information in here, then I can
remove the struct, but I don't want to be too hasty about it.

[reveman, 2014/10/27 17:14:43]

Ok, can you change the name to SolidColorAnalysis for now? I rather not have
unnecessarily abstract concepts just because we might want to extend it in the
future. Better to keep things as concrete and simple as possible and instead
rename it if we expand the usage.   

[vmpstr, 2014/10/27 18:46:40]

Done.

+
+  // Raster a subrect of this RasterSource into the given canvas. It is
+  // assumed that contents_scale has already been applied to this canvas.
+  // Writes the total number of pixels rasterized and the time spent
+  // rasterizing to the stats if the respective pointer is not nullptr.
+  virtual void RasterToBitmap(

[reveman, 2014/10/27 14:39:46]

hm, ToBitmap doesn't seem right. RasterToCanvas? or just Raster? or maybe even
ToCanvas? Or PlaybackToCanvas() to be consistent with the RasterBuffer API.

[vmpstr, 2014/10/27 15:55:11]

Changed to PlaybackToCanvas

+      SkCanvas* canvas,
+      const gfx::Rect& canvas_rect,
+      float contents_scale,
+      RenderingStatsInstrumentation* rendering_stats_instrumentation) const = 0;

[reveman, 2014/10/27 14:39:46]

Do we need to be passing this RenderingStatsInstrumentation pointer around?
Would be nice if that could just be handled by the implementation instead.

[vmpstr, 2014/10/27 15:55:11]

Good question. I'll talk to ernstm to see whether we actually need these stats
at all. Since most (all?) telemetry tests moved to tracing, I'm not sure we need
this. I've added a TODO for now until I find out for sure.

[reveman, 2014/10/27 17:14:43]

Acknowledged.

+
+  // Analyze to determine if the given rect at given scale is of solid color in
+  // this raster source.
+  virtual void AnalyzeInRect(
+      const gfx::Rect& content_rect,
+      float contents_scale,
+      Analysis* analysis,
+      RenderingStatsInstrumentation* rendering_stats_instrumentation) const = 0;
+
+  // Populate the given list with all SkPixelRefs that may overlap the given
+  // rect at given scale.
+  virtual void GatherPixelRefs(const gfx::Rect& content_rect,
+                               float contents_scale,
+                               std::vector<SkPixelRef*>* pixel_refs) const = 0;
+
+  // Return true iff this raster source can raster the given rect at given
+  // scale.
+  virtual bool RasterCoversRect(const gfx::Rect& content_rect,

[reveman, 2014/10/27 14:39:46]

CoversRect? RasterSource::CoversRect looks better to me than
RasterSource::RasterCoversRect.

[vmpstr, 2014/10/27 15:55:11]

Done.

+                                float contents_scale) const = 0;
+
+  // Return true iff this raster source would benefit from using distance
+  // fiend text.

[reveman, 2014/10/27 14:39:46]

typo: fiend

[vmpstr, 2014/10/27 15:55:11]

Done.

+  virtual bool SuitableForDistanceFieldText() const = 0;
+
+ protected:
+  friend class base::RefCountedThreadSafe<RasterSource>;
+
+  RasterSource() {}
+  virtual ~RasterSource() {}
+
+ private:
+  DISALLOW_COPY_AND_ASSIGN(RasterSource);
+};
+
+}  // namespace cc
+
+#endif  // CC_RESOURCES_RASTER_SOURCE_H_