add voice interaction shortcut.

components/arc/common/voice_interaction.mojom

Index: components/arc/common/voice_interaction.mojom
diff --git a/components/arc/common/voice_interaction.mojom b/components/arc/common/voice_interaction.mojom
new file mode 100644
index 0000000000000000000000000000000000000000..92029f495d1070c5d528a2b9982bc4a9b52ae72c
--- /dev/null
+++ b/components/arc/common/voice_interaction.mojom
@@ -0,0 +1,76 @@
+// Copyright 2017 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.
+//
+// Next MinVersion: 1
+
+module arc.mojom;
+
+// Represents view structure to be passed to ARC. The view
+// structure is synthesized from the AXStructure, which
+// is a simplified representation of the DOM tree. We
+// map each node of the AXStructure into a view element.
+struct VoiceInteractionStructure {
+  // Whether this is a valid response
+  bool success;

[Luis Héctor Chávez, 2017/03/13 17:39:34]

remove per the Optional<T> discussion.

[Muyuan, 2017/03/14 01:38:14]

OK.. after a second look, it seems that the mojo generator in Java does not
understand optional<T> yet?

[Luis Héctor Chávez, 2017/03/14 02:11:33]

it does: it generates a null Object. unfortunately java.util.Optional<T> is new
to Java 8, so we'll have to live with the current behavior until we drop support
for Marshmallow across the board.

[Muyuan, 2017/03/14 21:37:00]

Done.

+
+  // Geometry of the view in pixels
+  int32 x;

[Luis Héctor Chávez, 2017/03/13 17:39:35]

Can this be a
https://cs.chromium.org/chromium/src/ui/gfx/geometry/mojo/geometry.mojom?l=27
instead?

[Muyuan, 2017/03/14 01:38:14]

does geometry exists in Java side?

[Luis Héctor Chávez, 2017/03/14 02:11:33]

on further inspection, you should be able to use this instead:
https://cs.chromium.org/chromium/src/components/arc/common/screen_rect.mojom

it already maps to the correct types on both sides.

[Muyuan, 2017/03/14 21:37:00]

Done.

+  int32 y;
+  int32 width;
+  int32 height;
+
+  // Text of the view in utf-16 encoding
+  array<uint8> text;

[Luis Héctor Chávez, 2017/03/13 17:39:34]

Can this be https://cs.chromium.org/chromium/src/mojo/common/string16.mojom
instead? That way we can deal with base::string16 instead of having to manually
perform conversions.

[Muyuan, 2017/03/14 01:38:14]

same question as above..

[Luis Héctor Chávez, 2017/03/14 02:11:33]

you might need to do a similar trick as with ScreenRect to be able to map it to
an array<uint16>

+
+  // Text properties
+  float text_size;
+  int32 color;
+  int32 bgcolor;
+  bool bold;
+  bool italic;
+  bool underline;
+  bool line_through;
+
+  // Is the text selected
+  bool has_selection;
+  // Start and end unicode character position of the selection, inclusive.
+  int32 start_selection;
+  int32 end_selection;

[Luis Héctor Chávez, 2017/03/13 17:39:35]

Can you extract the start_selection and end_selection into a different structure
and add a nullable reference to such structure here to avoid the |has_selection|
bool?

[Muyuan, 2017/03/14 01:38:14]

So looks like optional is not implemented in the Java codegen.. Is there any
alternative to represent nullable object?

+
+  // Fake Android view class name of the element.  Each node is assigned
+  // a closest approximation of Android's views.
+  string class_name;
+
+  // Children of current node
+  array<VoiceInteractionStructure> children;
+};
+
+// Handles voice interaction queries from Android.
+// Next method ID: 2
+interface VoiceInteractionHost {
+  // Returns a screenshot of current focused window or empty bytes if
+  // no window is focused.
+  CaptureFocusedWindow@0() => (array<uint8> png_data);

[Luis Héctor Chávez, 2017/03/13 17:39:34]

What is the relationship between CaptureFocusedWindow and
GetVoiceInteractionStructure? Will they always be called in tandem? Can they be
collapsed to a single method to reduce raciness?

If they cannot be collapsed to a single method (possibly because they are called
from different instances), would it be possible to split the host such that each
class only implements the methods that the instance can call?

[Muyuan, 2017/03/14 01:38:14]

These are relatively independent by now.. One is called by system_server and the
other is called by ArcHomeService (which is not implemented atm). But the
problem here is that the future shape of both calls and additional
functionalities are not defined by UX.. so I don't know if separating them atm
is a good idea..

[Luis Héctor Chávez, 2017/03/14 02:11:33]

the problem is that substantially changing interfaces after the fact is
considerably more expensive. is there any way we can finalize the design before
committing to an interface?

+
+  // Returns view hierarchy of current window represented as
+  // VoiceInteractionStructure.
+  GetVoiceInteractionStructure@1() => (VoiceInteractionStructure structure);
+};
+
+// Connects with Android system server.
+// Next method ID:2
+interface VoiceInteractionCaptureInstance {
+  Init@0(VoiceInteractionHost host_ptr);
+
+  // Starts the voice interaction session in container.
+  StartVoiceInteractionSession@1();
+};
+
+// Connects with ArcHome.
+// Next method ID: 1
+interface VoiceInteractionArcHomeInstance {
+  Init@0(VoiceInteractionHost host_ptr);
+};
+

[Luis Héctor Chávez, 2017/03/13 17:41:02]

remove trailing empty lines.

[Muyuan, 2017/03/14 01:38:14]

Done.

+