Keep track of ChannelIDService in CookieStore

net/cookies/cookie_store.h

Index: net/cookies/cookie_store.h
diff --git a/net/cookies/cookie_store.h b/net/cookies/cookie_store.h
index c44139b0b87257001f2413ad89632ff8e31a8784..e9d463cc5fa4a2a98138fde3e9571cd2cdd9eff4 100644
--- a/net/cookies/cookie_store.h
+++ b/net/cookies/cookie_store.h
@@ -206,9 +206,12 @@ class NET_EXPORT CookieStore {
   // by some sort of persistence layer.
   // TODO(nharper): Remove this method once crbug.com/548423 has been closed.
   virtual bool IsEphemeral() = 0;
+  void SetChannelIDServiceID(int id);
+  int GetChannelIDServiceID();

[mmenke, 2016/03/28 17:54:00]

Do you plan to keep this, long term?  Why make the cookie store know about the
ChannelIDService, and not the other way around?  Not sure which is the right
dependency order, just wondering if you thought about it.

[nharper, 2016/03/29 00:30:20]

I'd like all of this debugging stuff to go away eventually (although that would
be problematic if we want to DCHECK when a mismatch happens).

The reason for having the cookie store know about the ChannelIDService is
because ChannelIDService already has an atomic sequence number for generating
these IDs - I believe I'd need to add something similar to CookieStore to have
this work the other way.

I didn't think very hard about the dependency order when I first did this, but I
think I'm convinced that this is the correct order. This set up allows for
multiple CookieStores to belong to the same ChannelIDService, which could be
fine (or could be problematic if deleting data from a CookieStore triggers
deletions from the ChannelIDService). If this is reversed, then a single
CookieStore could end up belonging to multiple ChannelIDServices, which would
always be wrong. The most robust way to ensure a 1-to-1 mapping would be to have
both know who the other is.

 
  protected:
   CookieStore();
+  int channel_id_service_id_;
 };
 
 }  // namespace net