Update build comment with results of analysis.

build/config/BUILDCONFIG.gn

 # Copyright (c) 2013 The Chromium Authors. All rights reserved.
 # Use of this source code is governed by a BSD-style license that can be
 # found in the LICENSE file.
 
 # =============================================================================
 # PLATFORM SELECTION
 # =============================================================================
 #
 # There are two main things to set: "os" and "cpu". The "toolchain" is the name
 # of the GN thing that encodes combinations of these things.
@@ skipping 625 matching lines @@
   if (is_component_build) {
     _component_mode = "shared_library"
   } else if (defined(invoker.static_component_type)) {
     assert(invoker.static_component_type == "static_library" ||
            invoker.static_component_type == "source_set")
     _component_mode = invoker.static_component_type
   } else if (is_android || !defined(invoker.sources)) {
     # When there are no sources defined, use a source set to avoid creating
     # an empty static library (which generally don't work).
     #
-    # TODO(brettw) remove the Android condition or comment why it needs to be
-    # kept after some analysis. Source sets vs. static libraries seem to be
-    # causing some performance differences. As part of the analysis for
-    # http://crbug.com/619593 we're testing source sets for components again.
+    # When we changed components to default from source sets to static
+    # libraries, an Android benchmark regressed slightly
+    # (https://crbug.com/619593). We don't have a good theory on why this might
+    # be since theoretically it should be the same. It could be something as
+    # silly as random code locality luck.
+    #
+    # There seems to be no build-time performance hit to using source sets on
+    # Android (the normal reason for defaulting to static libraries), so we
+    # make the default on Android to be source set.
+    #
+    # If it's been a long time since this was added and you're skeptical,
+    # please feel free to remove the Android exception and see if any
+    # benchmarks obviously regress. If not, it would be great to standardize
+    # with the rest of the platforms.
     _component_mode = "source_set"
   } else {
     _component_mode = "static_library"
   }
   target(_component_mode, target_name) {
     # Explicitly forward visibility, implicitly forward everything else.
     # Forwarding "*" doesn't recurse into nested scopes (to avoid copying all
     # globals into each template invocation), so won't pick up file-scoped
     # variables. Normally this isn't too bad, but visibility is commonly
     # defined at the file scope. Explicitly forwarding visibility and then
     # excluding it from the "*" set works around this problem.
     # See http://crbug.com/594610
     forward_variables_from(invoker, [ "visibility" ])
     forward_variables_from(invoker, "*", [ "visibility" ])
 
     # All shared libraries must have the sanitizer deps to properly link in
     # asan mode (this target will be empty in other cases).
     if (!defined(deps)) {
       deps = []
     }
     deps += [ "//build/config/sanitizers:deps" ]
   }
 }