[ChromeDriver] Select the main frame if a non-existant child frame is targeted.

chrome/test/chromedriver/devtools_client_impl.h

Index: chrome/test/chromedriver/devtools_client_impl.h
diff --git a/chrome/test/chromedriver/devtools_client_impl.h b/chrome/test/chromedriver/devtools_client_impl.h
index 4d03e78f2e7225fef3358c4936d081403ad9c0b5..2e3173f1f42c267416eddd328bf2c973bd8e244a 100644
--- a/chrome/test/chromedriver/devtools_client_impl.h
+++ b/chrome/test/chromedriver/devtools_client_impl.h
@@ -28,6 +28,10 @@ enum InspectorMessageType {
   kCommandResponseMessageType
 };
 
+enum InspectorErrorCode {
+  kContextIdNotFound = -32000,

[kkania, 2013/02/06 01:41:25]

-32000 just means generic server error.  devtools doesn't have very specific
error codes.  Here's our options:

1) add error codes
This will take a bit of work and you should probably consult with the devtools
guys first.  It might be hard to make this in before next week's branch.

2) compare message strings instead of the codes
This is a bit hacky, but probably fairly stable.  I think this one is "Execution
context with given id not found."  You can look at the webkit history to see if
this has ever changed.  If they do happen to change in the future, we can just
map multiple messages to the same error.  I think eventually we should fix this
though...

3) after receiving a generic error, check and see if the frame is detached; if
so, report error
This should work, but it doesn't seem that clean to me either.  This assumes
we'll receive the detach message before the command response, which probably
works now.

WHat do you think?

[craigdh, 2013/02/06 18:52:49]

Ah, that's unfortunate. I think for now the best choice is probably (2) as a
one-off, but we should reevaluate (1) if we find ourselves needing to watch for
more devtools errors in the future. I'm also not a big fan of (3), it seems
unnecessarily messy.

+};
+
 struct InspectorEvent {
   InspectorEvent();
   ~InspectorEvent();