Teach the mojo_shell --trace-startup flag to gather data from services

shell/desktop/main.cc

Index: shell/desktop/main.cc
diff --git a/shell/desktop/main.cc b/shell/desktop/main.cc
index 801574d0b9e99c876ff8fb70e9563c9e0b2142df..6139245aa70c35847f95b86bf2d10b37aa905203 100644
--- a/shell/desktop/main.cc
+++ b/shell/desktop/main.cc
@@ -91,14 +91,9 @@ int main(int argc, char** argv) {
 
   // We want the shell::Context to outlive the MessageLoop so that pipes are all
   // gracefully closed / error-out before we try to shut the Context down.
-  shell::Context shell_context;
+  shell::Context shell_context(&tracer);
   {
     base::MessageLoop message_loop;
-    if (!shell_context.Init()) {
-      Usage();
-      return 1;
-    }
-
     if (trace_startup) {
       message_loop.PostDelayedTask(
           FROM_HERE, base::Bind(&shell::Tracer::StopAndFlushToFile,
@@ -106,6 +101,11 @@ int main(int argc, char** argv) {
           base::TimeDelta::FromSeconds(5));
     }
 
+    if (!shell_context.Init()) {

[viettrungluu, 2015/04/27 17:04:11]

I don't know if the sketchy failure case is compatible with the post task above.
(I suppose it probably is, strictly speaking, but it only becomes even more
sketchy.)

[jamesr, 2015/04/27 23:46:57]

I think it's OK - even here the message_loop goes away before the shell_context
so if the task were to run somehow it wouldn't see any dead objects

+      Usage();
+      return 1;
+    }
+
     // The mojo_shell --args-for command-line switch is handled specially
     // because it can appear more than once. The base::CommandLine class
     // collapses multiple occurrences of the same switch.