First experimental implementation of the Clear-Site-Data header

content/public/browser/content_browser_client.h

Index: content/public/browser/content_browser_client.h
diff --git a/content/public/browser/content_browser_client.h b/content/public/browser/content_browser_client.h
index 6c7f8e6fb4a602db4b5edb4926b353f918085362..6088ecd8c4f7177a8e74a982f3380c64362bfe27 100644
--- a/content/public/browser/content_browser_client.h
+++ b/content/public/browser/content_browser_client.h
@@ -550,6 +550,14 @@ class CONTENT_EXPORT ContentBrowserClient {
   // Clears browser cookies.
   virtual void ClearCookies(RenderFrameHost* rfh) {}
 
+  // Clears |browser_context|'s data stored for the given |origin|.
+  // The datatypes to be removed are specified by |remove_cookies|,
+  // |remove_storage|, and |remove_cache|. Note that cookies will be removed
+  // for the entire eTLD+1 of |origin|.

[nasko, 2016/07/28 17:30:31]

Why would cookies and other types of data differ?

[msramek, 2016/08/01 16:03:19]

The proposal is to delete data for an origin as a security boundary. However,
when we started building the infrastructure to this change, it was pointed out
that "it will break the web". Websites generally use cookies from different
origins (e.g. mail.google.com from accounts.google.com); such partial deletion
would lead to unexpected behavior (inconsistent state), and it would also not
achieve the goal of clearing all places where the website could possibly store
data.

It's possible that we'll eventually decide to delete everything per origin
anyway, or that we'll switch to eTLD+1 for consistency. But for V1, we just want
to use what we have so that we can start experimenting in the real world.

+  virtual void ClearSiteData(
+      content::BrowserContext* browser_context, const url::Origin& origin,

[nasko, 2016/07/28 17:30:31]

style: Each parameter goes on a new line. Ideally, just run "git cl format" to
have automated tools take care of the formatting.

[msramek, 2016/08/01 16:03:19]

Done, also ran git cl format. (Some parts of this CL are definitely better
formatted now, but sometimes git cl format produces less readable results).

[nasko, 2016/08/03 20:53:13]

The main goal of git cl format is to remove any discussions on what looks more
or less readable, so please run it on all CLs. You remove a layer of cognitive
load from reviewers and lets them spend more mental energy on the things that
matter.

+      bool remove_cookies, bool remove_storage, bool remove_cache) {}
+
   // Returns the default download directory.
   // This can be called on any thread.
   virtual base::FilePath GetDefaultDownloadDirectory();