Added warning when URLRequestContextGetter leaks.

net/url_request/url_request_context_getter.cc

Index: net/url_request/url_request_context_getter.cc
diff --git a/net/url_request/url_request_context_getter.cc b/net/url_request/url_request_context_getter.cc
index db73fe3705f1e3016ed94993d757e3866c9b65cc..4cd9e3800f2510295a980a7d6622f3cec2394b91 100644
--- a/net/url_request/url_request_context_getter.cc
+++ b/net/url_request/url_request_context_getter.cc
@@ -22,7 +22,12 @@ void URLRequestContextGetter::OnDestruct() const {
     if (network_task_runner->BelongsToCurrentThread()) {
       delete this;
     } else {
-      network_task_runner->DeleteSoon(FROM_HERE, this);
+      if (!network_task_runner->DeleteSoon(FROM_HERE, this)) {
+        // Can't force-delete the object here, because some derived classes

[eroman, 2012/08/15 18:00:44]

Please change the language to not reference chrome concepts (like "IO thread"
and "ShellURLRequestContextGetter").

This may seem silly, but conceptually the net/* directory should be a separable
module with no dependencies on Chrome, even though it is mainly intended to be
embedded into Chrome.

[Andrew T Wilson (Slow), 2012/08/15 18:07:17]

Ah, good point. I'll do this.

+        // (like ShellURLRequestContextGetter) may only be deleted on the IO
+        // thread.
+        DLOG(WARNING) << "URLRequestContextGetter leaking due to no IO thread.";

[eroman, 2012/08/15 18:00:44]

How about making this simply LOG(WARNING)?

[Andrew T Wilson (Slow), 2012/08/15 18:07:17]

Since this really only impacts unit tests, I figured it wasn't worth increasing
the size of the release version of chrome by including a warning.

+      }
     }
   }
   // If no IO message loop proxy was available, we will just leak memory.