Xorg+dummy: set resolution to first size in list

remoting/host/linux/linux_me2me_host.py

Index: remoting/host/linux/linux_me2me_host.py
diff --git a/remoting/host/linux/linux_me2me_host.py b/remoting/host/linux/linux_me2me_host.py
index 6050ee8546aedc16514a1d86bec1566f52193038..b77c367f56620f9c6d2ef5030d9a87614c2c2ff1 100755
--- a/remoting/host/linux/linux_me2me_host.py
+++ b/remoting/host/linux/linux_me2me_host.py
@@ -518,6 +518,7 @@ class Desktop:
     self.child_env["XAUTHORITY"] = x_auth_file
     devnull = open(os.devnull, "r+")
     display = self.get_unused_display_number()
+    use_xorg = USE_XORG_ENV_VAR in os.environ
 
     # Run "xauth add" with |child_env| so that it modifies the same XAUTHORITY
     # file which will be used for the X session.
@@ -534,7 +535,7 @@ class Desktop:
         x_session[1] == "/usr/bin/gnome-session --session=ubuntu-2d"):
       extra_x_args.extend(["-extension", "Composite"])
 
-    if USE_XORG_ENV_VAR in os.environ:
+    if use_xorg:
       self._launch_xorg(display, x_auth_file, extra_x_args)
     else:
       self._launch_xvfb(display, x_auth_file, extra_x_args)
@@ -574,6 +575,16 @@ class Desktop:
     if retcode != 0:
       logging.error("Failed to set XKB to 'evdev'")
 
+    # Xorg+dummy will default to a resolution that is the maximum of all
+    # specified widths by the maximum of all the specified heights (which might
+    # even be a mode that wasn't specified by the user). Instead, we want to
+    # mimic the behavior we have with Xvfb, where the first size in the list is
+    # used as the initial size. Note: The session's settings daemon might change
+    # to a different resolution when it starts.
+    if use_xorg:
+      subprocess.call(["xrandr", "--size", "%dx%d" % self.sizes[0]],
+                      env=self.child_env)

[Lambros, 2016/05/17 00:08:14]

Should redirect stdout/stderr (as in the code below).

[rkjnsn, 2016/05/17 21:58:45]

Is there a reason we wouldn't want an error message here to show up in the log
file?

[Lambros, 2016/05/18 00:13:49]

I suppressed the errors originally because Xvfb-randr always whines about some
"CRT" problem, which is a false alarm.

You could show the errors, I suppose. There is already lots of noise in the
session logs anyway. But be consistent with all the xrandr commands (feel free
to punt to a separate CL).

+
     if not self.server_supports_exact_resize:

[Lambros, 2016/05/17 00:08:14]

You currently set this to False for Xorg+dummy, which means this will
early-return. But the code you added above is the same as line 608 below.

Maybe refactor this a little? 'server_supports_exact_resize' feels like the
wrong thing to test. We should be testing the level of XRANDR support. For
example, xrandr -s will work just fine with Xorg+dummy, but the
--newmode/--addmode commands might not work properly?

[rkjnsn, 2016/05/17 21:58:45]

I agree that ‘server_supports_exact_resize’ probably isn't the right test in the
long run, but testing the level of RandR would definitely be wrong. Even once we
have a version of dummy that supports exact resize, we will still be providing
the initial resolutions through the config file, and won't want to do the
--newmode/--addmode commands. Maybe we should have a
"self.server_supports_randr" flag that controls setting the initial size,
setting DPI, and spawning the watch-resolution logic, and separate logic that
only runs in the Xvfb-randr case to handle adding resolutions.

[Lambros, 2016/05/18 00:13:49]

Sounds good.

       return