Deflake WidgetInputMethodInteractiveTest.TwoTopWindows on Linux

I was able to repro the flake with ~1% probability and traced.

It was flaky because earlier window server messages were still pending
when the test started running checks. So the first `ActivateSync` would
actually break out of the message loop wait too soon.

BUG=484836

Review URL: https://codereview.chromium.org/1130683004

Cr-Commit-Position: refs/heads/master@{#329267}

1 files changed, 6 insertions, 4 deletions

diff --git a/ui/views/widget/widget_interactive_uitest.cc b/ui/views/widget/widget_interactive_uitest.cc
index f9d6605..5c27773 100644
--- a/ui/views/widget/widget_interactive_uitest.cc
+++ b/ui/views/widget/widget_interactive_uitest.cc
@@ -1582,8 +1582,7 @@ TEST_F(WidgetInputMethodInteractiveTest, TwoWindows) {
 
 // Test input method focus changes affected by focus changes cross 2 top
 // windows.
-// Flaky: https://crbug.com/484836
-TEST_F(WidgetInputMethodInteractiveTest, DISABLED_TwoTopWindows) {
+TEST_F(WidgetInputMethodInteractiveTest, TwoTopWindows) {
   Widget* widget1 = CreateWidget();
   Widget* widget2 = CreateWidget();
   Textfield* textfield1 = new Textfield;
@@ -1591,8 +1590,11 @@ TEST_F(WidgetInputMethodInteractiveTest, DISABLED_TwoTopWindows) {
   textfield2->SetTextInputType(ui::TEXT_INPUT_TYPE_PASSWORD);
   widget1->GetRootView()->AddChildView(textfield1);
   widget2->GetRootView()->AddChildView(textfield2);
-  widget1->Show();
-  widget2->Show();
+
+  // Do the initial shows synchronously. Otherwise, on X11, the window server
+  // messages may be interleaved with the activation requests below.
+  ShowSync(widget1);
+  ShowSync(widget2);
 
   textfield1->RequestFocus();
   textfield2->RequestFocus();