More strict restrictions for content -> content/shell dependencies.

content/common/DEPS

Index: content/common/DEPS
diff --git a/content/common/DEPS b/content/common/DEPS
index 89bdfb6c05a6a35b17abefe804a7f104f520c4d0..f2e00f19aff1f6cfb6be09012ad34781689ebf08 100644
--- a/content/common/DEPS
+++ b/content/common/DEPS
@@ -66,8 +66,13 @@ include_rules = [
 ]
 
 specific_include_rules = {
-# Java bridge code passes NPAPI types to the browser process. Crazy!
+  # Java bridge code passes NPAPI types to the browser process. Crazy!
   "java_bridge_messages\.h": [

[jochen (gone - plz use gerrit), 2016/03/09 14:57:19]

that file doesn't exist anymore, right?

[Łukasz Anforowicz, 2016/03/09 21:47:42]

Done - removed.

buildtools/checkdeps/checkdeps.py content # continues to pass.

I did find content/common/gin_java_bridge_messages.h, but 1) it looks to be
something different + 2) it doesn't include anything from content/child.

     "+content/child"
+  ],
+
+  # TODO(nasko): https://crbug.com/357747: Should remove swapped out altogether.

[jochen (gone - plz use gerrit), 2016/03/09 14:57:19]

the CL that introduced this fixes a single layout test. can we just skip the
layout test instead?

[Łukasz Anforowicz, 2016/03/09 21:47:42]

I'd rather leave this here - currently (now that we've OOPIF-ied a significant
chunk of layout test harness) removing the ShellViewHostMsg_PrintMessage
exclusion in swapped_out_message.cc leads to 222 failures (text/diffs) under
http/tests when layout tests are run with --site-per-process.  For example -
this breaks all tests that expect the actual output to contain lines from
|alert('blah')| kind of calls.

I think we also have a tentative plan for getting rid of the swapped out
exclusions even in case swapped out cannot be yanked out soon - we can send
messages through a local frame, rather than sending them via (swapped out)
render view.  Still - this is a tentative plan - before we commit to it we
should have a clearer picture of which pieces of BlinkTestRunner (and
BlinkTestController) are per-view vs per-process vs per-frame and only after
having that picture change how messages are sent around.  This might also be an
interesting mojofication experiment (although I am very skeptical about the
absence of ordering guarantees for navigation-vs-layouttest messages).

I guess the last paragraph is a long-winded way of trying to say "It will really
be temporary.  Pinky-promise!".  Hmmmm... :-)  OTOH, I am not sure if we have
good alternatives and the layering violation is narrow and won't grow (after
landing the DEPS CL).

[Łukasz Anforowicz, 2016/03/22 17:29:31]

Done - it turns out that there already is a public API for filtering swapped out
messages (ContentClient::CanSendWhileSwappedOut) - layout tests should simply
reuse this.

+  "swapped_out_messages\.cc": [
+    "+content/shell/common/shell_messages.h"
   ]
 }