Reland of Make it possible for tests to set up trace event handlers.

webrtc/BUILD.gn

Index: webrtc/BUILD.gn
diff --git a/webrtc/BUILD.gn b/webrtc/BUILD.gn
index 58b5f5e74004efdf86ca154e3e838d678faeeb16..6d2c0e426c1d3f984a783d56e37b23acfb1595d6 100644
--- a/webrtc/BUILD.gn
+++ b/webrtc/BUILD.gn
@@ -26,8 +26,19 @@ config("common_inherited_config") {
   if (build_with_mozilla) {
     defines += [ "WEBRTC_MOZILLA_BUILD" ]
   }
+  # Some tests need to declare their own trace event handlers. If this define is
+  # not set, the first time TRACE_EVENT_* is called it will store the return
+  # value for the current handler in an static variable, so that subsequent
+  # changes to the handler for that TRACE_EVENT_* will be ignored.
+  # So when tests are included, we set this define, making it possible to use
+  # different event handlers in different tests.
+  if (rtc_include_tests) {
+    defines += [ "WEBRTC_NON_STATIC_TRACE_EVENT_HANDLERS=1" ]
+  } else {
+    defines += [ "WEBRTC_NON_STATIC_TRACE_EVENT_HANDLERS=0" ]
+  }
   if (build_with_chromium) {
-    defines = [
+    defines += [

[mbonadei, 2017/08/17 14:02:50]

Oh... This was a well hidden bug. :)

I thought that GN should complain if you try to override an already defined
variable.

[ehmaldonado_webrtc, 2017/08/17 14:05:13]

It did, hence the failure to generate build files in the Chromium bot. The
WebRTC bots didn't catch this because build_with_chromium is false.

[mbonadei, 2017/08/17 14:08:34]

I see... Yeah, I didn't look at the failure. :)

       # TODO(kjellander): Cleanup unused ones and move defines closer to
       # the source when webrtc:4256 is completed.
       "FEATURE_ENABLE_VOICEMAIL",