Continue browser startup after killing a hung browser instance.

chrome/browser/process_singleton_win.cc

Index: chrome/browser/process_singleton_win.cc
diff --git a/chrome/browser/process_singleton_win.cc b/chrome/browser/process_singleton_win.cc
index 75918cd646d30acc63b9ec69397b038600f8caa0..453824d8ffff3ea2eafaef1bd38af5364dba6c33 100644
--- a/chrome/browser/process_singleton_win.cc
+++ b/chrome/browser/process_singleton_win.cc
@@ -278,7 +278,8 @@ ProcessSingleton::NotifyResult ProcessSingleton::NotifyOtherProcess() {
   // Time to take action. Kill the browser process.
   process.Terminate(content::RESULT_CODE_HUNG, true);
   remote_window_ = NULL;
-  return PROCESS_NONE;
+  // GCOMANICI return PROCESS_HUNG;
+  return PROCESS_HUNG;
 }
 
 ProcessSingleton::NotifyResult
@@ -286,9 +287,22 @@ ProcessSingleton::NotifyOtherProcessOrCreate() {
   ProcessSingleton::NotifyResult result = PROCESS_NONE;
   if (!Create()) {
     result = NotifyOtherProcess();
+    if (result == PROCESS_HUNG) {
+      // The Chrome instance that usses the main profile directory was
+      // detected to be hung and the user confirmed its termination.
+      // Current process should have a secont attempt to create the profile
+      if (!Create()) {

[grt (UTC plus 2), 2016/08/26 09:26:43]

nit: omit braces for conditionals where the condition and the body are each one
line.

[gcomanici, 2016/08/26 17:26:59]

Acknowledged.

+        result = NotifyOtherProcess();

[grt (UTC plus 2), 2016/08/26 09:26:43]

instead of duplicating the calls to create and notify, i think it's cleaner to
use a loop:

  for (int i = 0; i < 2; ++i) {
    if (Create())
      return PROCESS_NONE;  // This is the single browser process.
    NotifyResult result = NotifyOtherProcess();
    if (result == PROCESS_NOTIFIED || result == LOCK_ERROR) {
      // The single browser process was notified, the user chose not to
      // terminate a hung browser, or the lock file could not be created.
      // Nothing more to do.
      return result;
    }
    DCHECK_EQ(PROCESS_NONE, result);
    // The process could not be notified for some reason, or it was hung and
    // terminated. Retry once if this is the first time; otherwise, fall through
    // to report that the process must exit because the profile is in use.
  }
  return PROFILE_IN_USE;

[gcomanici, 2016/08/26 17:26:59]

Done.

+      }
+    }
     if (result == PROCESS_NONE)
       result = PROFILE_IN_USE;
-  } else {
+    if (result == PROCESS_HUNG) {
+      // This only happens if the second attempt at Create() was succesful.
+      result = PROCESS_NONE;
+    }
+  }
+  if (result == PROCESS_NONE) {
     g_browser_process->platform_part()->PlatformSpecificCommandLineProcessing(

[grt (UTC plus 2), 2016/08/26 09:26:43]

looking at the design in here a bit more, it seems to me that the intent is to
not have dependencies on Chrome: this is what ChromeProcessSingleton is for. i
suggest moving this call into
ChromeProcessSingleton::NotifyOtherProcessOrCreate() (and clean up the #includes
in this file).

[gcomanici, 2016/08/26 17:26:59]

The new patch has this call in chrome_browser_main.cc. I looked into
ChromeProcessSingleton::NotifyOtherProcessOrCreate() and there is no need to
have the call inside NotifyOtherProcessOrCreate(). It would only create more
imports in chrome_process_singleton.cc.

         *base::CommandLine::ForCurrentProcess());
   }