Use FrameIsAd to decide whether to isolate a frame in TopDocumentIsolation mode.

content/DEPS

 # Do NOT add chrome to the list below. We shouldn't be including files
 # from src/chrome in src/content.
 include_rules = [
   # The subdirectories in content/ will manually allow their own include
   # directories in content/ so we disallow all of them.
   "-content",
   "+content/app/resources/grit/content_resources.h",
   "+content/common",
   "+content/grit",
   "+content/public/common",
   "+content/public/test",
   "+content/test",
 
   "+blink/public/resources/grit",
 
   "+cc",
   "-cc/blink",
 
   "-components",
   # Content can depend on components that are:
   #   1) related to the implementation of the web platform
   #   2) shared code between third_party/WebKit and content
   # It should not depend on chrome features or implementation details, i.e. the
   # original components/ directories which was code split out from chrome/ to be
   # shared with iOS. This includes, but isn't limited to, browser features such
   # as autofill or extensions, and chrome implementation details such as
   # settings, packaging details, installation or crash reporting.
+  "+components/variations",

[Alexei Svitkine (slow), 2017/07/07 16:55:05]

I do wonder if we shouldn't perhaps restrict this more - i.e. only allow
specific parts of the component to be includable? I suspect it should only be a
handful of files, so maybe we could just list them here with a comment
mentioning that only that functionality is supported to be used from content.

[Łukasz Anforowicz, 2017/07/07 20:41:42]

Good point.  I've made changes that I think should address this concern.  I've
included a long set of notes below, but after making the changes I've made, I am
not sure if they are that interesting.

These are the includes I see after the CL:

$ git grep include.*components.variations -- content/
content/browser/memory/memory_condition_observer.cc:#include
"components/variations/variations_associated_data.h"
content/browser/renderer_host/render_view_host_impl.cc:#include
"components/variations/variations_associated_data.h"
content/ppapi_plugin/ppapi_thread.h:#include
"components/variations/child_process_field_trial_syncer.h"
content/public/test/test_utils.cc:#include
"components/variations/variations_switches.h"
content/renderer/render_thread_impl.h:#include
"components/variations/child_process_field_trial_syncer.h"

I see the following groups of related functionality:

- components/variations/child_process_field_trial_syncer.h
(variations::ChildProcessFieldTrialSyncer)

  - For now I moved it back to content/renderer/DEPS and
    content/ppapi_plugin/DEPS
    and made it specific to this particular header.
  
  - This is a field in both content::RenderThreadImpl and content::PpapiThread.

  - Maybe this field should be moved to content::ChildThreadImpl?

    - Such move would help hide/encapsulate the dependency on
      components/variations/child_process_field_trial_syncer.h

    - This might require moving field_trial_syncer_.InitFieldTrialObserving(...)
call
      from RenderThreadImpl::Init to the constructor of ChildThreadImpl.

    - This might require having ChildThreadImpl's constructor pass
      base::CommandLine::ForCurrentProcess as an argument for
      field_trial_syncer_.InitFieldTrialObserving
      (like RenderThreadImpl::Init does today;  a little bit unlike
      PpapiThread does today [although ultimately the value of
      MainFunctionParams::command_line also comes from
      base::CommandLine::ForCurrentProcess - called in
      ContentMainRunnerImpl::Run]

    - I am not sure what to do about RenderThreadImpl::SetFieldTrialGroup.
      Maybe this can also move to ChildThreadImpl.

    - I wonder if not having SetFieldTrialGroup on PpapiThread might indicate
      a bug in propagation of field trial parameters to the ppapi process.
  

- components/variations/variations_associated_data.h (GetVariations...
functions)

  - This seems unused in memory_condition_observer.cc
    I've removed the include here.

  - This seems removable in render_view_host_impl.cc.
    OTOH, https://crbug.com/681160 is fixed and there is no active bug to track
the removal...
    FWIW, I opened a new bug and moved this to "specific_include_rules"
    in content/browser/renderer_host/DEPS.

- components/variations/variations_switches.h

  - This is the addition in this CL needed for testing.

  - As you suggested I am trying to use VariationsParamManager instead.

  - I've added a narrow dependency from //content's browser tests and test
utils.
    

[Alexei Svitkine (slow), 2017/07/10 18:46:24]

That SGTM for now. Maybe we can file a crbug for the suggestions you have below?
That SGTM. My mental model is we can add that include for now and it can be
removed over time.

[Łukasz Anforowicz, 2017/07/10 22:17:39]

Done - https://crbug.com/740726.
Actually, I was able to already do it after rebasing - servolk@ has already
kindly fixed https://crbug.com/740174 :-)

 
   "+crypto",
   "+grit/blink_resources.h",
   "+grit/content_strings.h",
 
   "+dbus",
   "+gpu",
   "+media",
   "+mojo/common",
   "+mojo/edk/embedder",
@@ skipping 80 matching lines @@
   "+jni",
 ]
 
 # content -> content/shell dependency is not allowed, except for browser tests.
 specific_include_rules = {
   ".*_browsertest[a-z_]*\.(cc|h)": [
     "+content/shell/browser",
     "+content/shell/common",
   ],
 }