Use the physical-pixel space for native IME on linux platform.

ui/base/ime/input_method_win.cc

Index: ui/base/ime/input_method_win.cc
diff --git a/ui/base/ime/input_method_win.cc b/ui/base/ime/input_method_win.cc
index 49efa46632113f4ed070bb20fea853d86bc49869..dbca1174fad9b752b9e3721e338cef12673caf4d 100644
--- a/ui/base/ime/input_method_win.cc
+++ b/ui/base/ime/input_method_win.cc
@@ -16,11 +16,11 @@
 #include "ui/base/ime/text_input_client.h"
 #include "ui/base/ime/win/tsf_input_scope.h"
 #include "ui/base/ui_base_switches.h"
-#include "ui/display/win/screen_win.h"
 #include "ui/events/event.h"
 #include "ui/events/event_constants.h"
 #include "ui/events/event_utils.h"
 #include "ui/events/keycodes/keyboard_codes.h"
+#include "ui/gfx/geometry/rect.h"
 #include "ui/gfx/win/hwnd_util.h"
 
 namespace {
@@ -233,12 +233,12 @@ void InputMethodWin::OnCaretBoundsChanged(const TextInputClient* client) {
 
   // The current text input type should not be NONE if |client| is focused.
   DCHECK(!IsTextInputTypeNone());
-  // Tentatively assume that the returned value is DIP (Density Independent
-  // Pixel). See the comment in text_input_client.h and http://crbug.com/360334.
+  // Windows IME uses the non-dip coordinates.
+  // See the comment in text_input_client.h and http://crbug.com/360334.
   const gfx::Rect dip_screen_bounds(GetTextInputClient()->GetCaretBounds());
   const gfx::Rect screen_bounds =
-      display::win::ScreenWin::DIPToScreenRect(toplevel_window_handle_,

[sadrul, 2016/12/22 01:30:22]

Does the old code not do the right thing, because it looks like that too tries
to convert from DIP to pixel-space?

[Shu Chen, 2016/12/22 03:02:27]

The old code does the right thing but I think it would be better to use the
consistent way to do the coordinates transform cross platforms (win/linux).

 

-                                               dip_screen_bounds);
+      gfx::ScaleToEnclosingRect(
+          dip_screen_bounds, GetTextInputClient()->GetScaleFactor());
 
   HWND attached_window = toplevel_window_handle_;
   // TODO(ime): see comment in TextInputClient::GetCaretBounds(), this
@@ -630,9 +630,6 @@ LRESULT InputMethodWin::OnQueryCharPosition(IMECHARPOSITION* char_positon) {
   if (!client)
     return 0;
 
-  // Tentatively assume that the returned value from |client| is DIP (Density
-  // Independent Pixel). See the comment in text_input_client.h and
-  // http://crbug.com/360334.
   gfx::Rect dip_rect;
   if (client->HasCompositionText()) {
     if (!client->GetCompositionCharacterBounds(char_positon->dwCharPos,
@@ -646,9 +643,10 @@ LRESULT InputMethodWin::OnQueryCharPosition(IMECHARPOSITION* char_positon) {
       return 0;
     dip_rect = client->GetCaretBounds();
   }
+  // Windows IME uses the non-dip coordinates.
+  // See the comment in text_input_client.h and http://crbug.com/360334.
   const gfx::Rect rect =
-      display::win::ScreenWin::DIPToScreenRect(toplevel_window_handle_,
-                                               dip_rect);
+      gfx::ScaleToEnclosingRect(dip_rect, client->GetScaleFactor());

[sadrul, 2016/12/22 01:30:22]

Can you use ConvertRectToDIP/ConvertRectToPixel instead?

https://cs.chromium.org/chromium/src/ui/compositor/dip_util.h?q=ConvertRectTo...

[Shu Chen, 2016/12/22 03:02:27]

I don't think so, because that would require a Layer instance which doesn't
available in ui/base/ime space.

 
   char_positon->pt.x = rect.x();
   char_positon->pt.y = rect.y();