When resizing-to-fit, prefer to match desktop rotation if possible.

remoting/host/desktop_resizer_win.cc

Index: remoting/host/desktop_resizer_win.cc
diff --git a/remoting/host/desktop_resizer_win.cc b/remoting/host/desktop_resizer_win.cc
index fff0355dd64d22c75df4567af313f58b06241097..8f74cc03eb018d78f88da7348d7742d79eb9ed14 100644
--- a/remoting/host/desktop_resizer_win.cc
+++ b/remoting/host/desktop_resizer_win.cc
@@ -33,7 +33,7 @@ class DesktopResizerWin : public DesktopResizer {
   static bool IsResizeSupported();
 
   // Calls EnumDisplaySettingsEx() for the primary monitor.
-  // Returns a DEVMODE with no fields if |mode_number| does not exist.
+  // Returns false if |mode_number| does not exist.
   static bool GetPrimaryDisplayMode(
       DWORD mode_number, DWORD flags, DEVMODE* mode);
 
@@ -91,14 +91,28 @@ std::list<SkISize> DesktopResizerWin::GetSupportedSizes(
     if (candidate_mode.dmBitsPerPel != current_mode.dmBitsPerPel)
       continue;
 
-    // If there is an existing mode of the same dimensions, prefer the
-    // one with the higher frequency.
+    // If there is an existing mode of the same dimensions, choose the
+    // closest match to the current mode.
     SkISize candidate_size = GetModeSize(candidate_mode);
     if (modes_.count(candidate_size) != 0) {
       DEVMODE existing_mode = modes_[candidate_size];

[Jamie, 2013/02/13 18:21:01]

Would best_mode_for_size be a better name for this variable? I spent a few
minutes trying to work out what it was for.

[Wez, 2013/02/13 22:44:55]

You mean modes_ -> best_mode_for_size_?

[Jamie, 2013/02/14 00:10:34]

I meant best_mode_for_size instead of existing_mode, but actually I like your
suggestion better.

-      if (existing_mode.dmDisplayFrequency >
-          candidate_mode.dmDisplayFrequency)
+
+      // Skip this mode if its rotation is a worse match for the current mode.
+      if ((candidate_mode.dmDisplayOrientation !=
+           current_mode.dmDisplayOrientation) &&
+          (existing_mode.dmDisplayOrientation ==
+           current_mode.dmDisplayOrientation)) {

[Jamie, 2013/02/13 18:21:01]

This will never pick a portrait mode if the physical display is landscape,
correct? I'm not sure that's the right choice for Me2Me, especially if the
display is curtained. I think it deserves a comment at least, explaining that
this is a design choice.

[Wez, 2013/02/13 22:44:55]

No.  This logic applies iff the mode we're enumerating, and the previous
best-match mode have the same dimensions.  If the new mode's orientation doesn't
match the current mode, but the best-match mode's does, then we prefer the
best-match mode still - i.e. we prefer matching the orientation over matching
the refresh rate.

I'll rephrase this as a single comment explaining that prioritisation across
these blocks.

[Jamie, 2013/02/14 00:10:34]

Oh, I see. I had it in mind that the API returned the unrotated width and
height, but that wouldn't be very useful... In that case, this is fine.

+        continue;
+      }
+
+      // Skip this mode if its frequency does not match, and the existing
+      // choice has a higher frequency.
+      if ((candidate_mode.dmDisplayFrequency !=
+           current_mode.dmDisplayFrequency) &&
+          (existing_mode.dmDisplayFrequency >=
+           candidate_mode.dmDisplayFrequency)) {
         continue;
+      }
     } else {
       // If we haven't seen this size before, add it to those we return.
       sizes.push_back(candidate_size);