UMA metrics for use count of wheel and touch scrolls.

cc/trees/layer_tree_host_common.h

 // Copyright 2011 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_TREES_LAYER_TREE_HOST_COMMON_H_
 #define CC_TREES_LAYER_TREE_HOST_COMMON_H_
 
 #include <stddef.h>
 
 #include <limits>
@@ skipping 159 matching lines @@
   // current page scale, the two must be committed at the same time to prevent
   // clamping.
   LayerTreeHostCommon::ScrollUpdateInfo inner_viewport_scroll;
 
   std::vector<LayerTreeHostCommon::ScrollUpdateInfo> scrolls;
   float page_scale_delta;
   gfx::Vector2dF elastic_overscroll_delta;
   float top_controls_delta;
   std::vector<LayerTreeHostCommon::ScrollbarsUpdateInfo> scrollbars;
   std::vector<std::unique_ptr<SwapPromise>> swap_promises;
+  uint32_t scroll_source_info;

[danakj, 2017/03/10 20:52:48]

Why did you go with a bitmask here instead of 2 bools? You can make them 1 bit
large each here.

[sahel, 2017/03/13 20:13:43]

If other types of scrolling (e.g. keyboard) are handled by the compositor thread
in future (there are some plans for keyboard events generating gesture events)
this implementation is extendable.

In terms of struct size I didn't think it's critical because the struct already
has vectors. I changed it to uint8_32 though, if it's still not good enough I
can change it to two boolean values.

[danakj, 2017/03/13 20:16:23]

I don't see how a bitflag is any more extendable than a set of bools. Maybe you
can explain?
I was wondering why you specify a size at all rather than the size you did. The
compiler will pick a reasonable size for an enum.

[sahel, 2017/03/13 21:03:52]

what I meant is: if I want to add other sources of scrolling, with having one
flag per source type I will keep updating the struct. On the other hand with an
enum, all source info is wrapped into one field regardless of the number of
resources and I won't need to change the struct and its populating when I need
to add a new resource. Anyways, I can definitely change it to one flag per type
if you prefer them.  

I didn't know that compilers do that. Thanks for explaining. 

[danakj, 2017/03/13 22:29:15]

Optimize for the reader, not the writer. A uint32_t will explain much less about
what you're reading here than a couple bools will. So yeah, I prefer that.

[sahel, 2017/03/16 12:58:26]

Done.

 
  private:
   DISALLOW_COPY_AND_ASSIGN(ScrollAndScaleSet);
 };
 
 template <typename Function>
 void LayerTreeHostCommon::CallFunctionForEveryLayer(LayerTreeHost* host,
                                                     const Function& function) {
   for (auto* layer : *host) {
     function(layer);
@@ skipping 12 matching lines @@
     function(tree_impl->LayerById(id));
   }
 }
 
 CC_EXPORT PropertyTrees* GetPropertyTrees(Layer* layer);
 CC_EXPORT PropertyTrees* GetPropertyTrees(LayerImpl* layer);
 
 }  // namespace cc
 
 #endif  // CC_TREES_LAYER_TREE_HOST_COMMON_H_