Add an OriginFilterBuilder class for [white|black]listing origins.

chrome/browser/browsing_data/origin_filter_builder.h

Index: chrome/browser/browsing_data/origin_filter_builder.h
diff --git a/chrome/browser/browsing_data/origin_filter_builder.h b/chrome/browser/browsing_data/origin_filter_builder.h
new file mode 100644
index 0000000000000000000000000000000000000000..6712afeb99704db95a0cf9ff80794e5fd1a630d2
--- /dev/null
+++ b/chrome/browser/browsing_data/origin_filter_builder.h
@@ -0,0 +1,72 @@
+// Copyright 2016 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.
+
+#ifndef CHROME_BROWSER_BROWSING_DATA_ORIGIN_FILTER_BUILDER_H_
+#define CHROME_BROWSER_BROWSING_DATA_ORIGIN_FILTER_BUILDER_H_
+
+#include <ostream>
+#include <set>
+#include <vector>
+
+#include "base/callback.h"
+#include "url/gurl.h"
+#include "url/origin.h"
+
+// A class that constructs URL deletion filters (represented as GURL->bool
+// predicates) that match certain origins.
+class OriginFilterBuilder {
+ public:
+  enum Mode {
+    WHITELIST,
+    BLACKLIST
+  };
+
+  // Constructs a filter with the given |mode| - whitelist or blacklist.
+  explicit OriginFilterBuilder(Mode mode);
+
+  ~OriginFilterBuilder();
+
+  // Adds the |origin| to the (white- or black-) list, and returns itself
+  // for method call chaining.
+  OriginFilterBuilder* AddOrigin(const url::Origin& origin);

[brettw, 2016/02/17 20:34:24]

Thanks for the update on the builder pattern, I like this better than the old
way.

[msramek, 2016/02/18 12:48:43]

Acknowledged.

+
+  // Change the |mode| of the filter.
+  OriginFilterBuilder* ChangeMode(Mode mode);

[Mike West, 2016/02/17 15:22:10]

Nit: s/ChangeMode/SetMode/

[brettw, 2016/02/17 20:34:24]

This, plus void return value. I notice you do this to let callers do
builder.SetMode(...)->Bar(). In some cases this pattern can be useful. But this
is pretty unusual in Chrome, has potentially confusing memory management
semantics, and the way you use this in the patch, it's the same number of lines
to do
  builder.SetMode();
  builder.AddOrigin();
on separate lines.

[msramek, 2016/02/18 12:48:43]

Done. Didn't seem unusual to me, as as I have recently worked with
ContentSettingsPattern::Builder, which does that :)

+
+  // Builds a filter that matches URLs whose origins are in the whitelist,
+  // or aren't in the blacklist. Once a filter is built, it is no longer
+  // possible to change the builder parameters (origin list and mode).
+  base::Callback<bool(const GURL&)> BuildSameOriginFilter();
+
+  // Build a filter that matches URLs whose origins, or origins obtained by
+  // replacing the host with any superdomain, are listed in the whitelist,
+  // or are not listed in the blacklist. Once a filter is built, it is no longer
+  // possible to change the builder parameters (origin list and mode).

[Mike West, 2016/02/17 15:22:10]

Can you add some comments about lifetimes here and `BuildSameOriginFilter()`? As
ttr314@ noted, that's actually a bit complicated.

[msramek, 2016/02/18 12:48:43]

My original thought here was that BrowsingDataRemover blocks until the deletion
finishes anyway, so there won't be a problem with lifetimes.

However, in Timo's CL for password manager, this caused some troubles in tests.
We discussed this in a separate thread, and my proposal (which I implemented in
this last patchset) is to make a copy of origin_list_ and pass it as Owned()
instead of Unretained(). This means that the callback will keep its list of URLs
as long as they're necessary, and there will be no surprises w.r.t. lifetime.

Since we established that the number of origins typically shouldn't be large
(which is why we can get away with set instead of e.g. trie), we can also reckon
that this approach will still be fast and not take too much extra memory.

It also allows me to remove the DCHECKS saying "you cannot modify the builder
after building once", because now each built filter will make its own copy of
the data.

And finally, it's weird for a builder to have to stay alive after building. I
think the current state is much better.

+  base::Callback<bool(const GURL&)> BuildDomainFilter();

[Mike West, 2016/02/17 15:22:10]

HSTS calls these "congruent" and "superdomain" matches. Perhaps reusing that
terminology here could be useful:
`BuildCongruentFilter`/`BuildSuperdomainFilter`?

[brettw, 2016/02/17 20:34:24]

I've never heard of those, and I personally prefer the patches current naming.

[Mike West, 2016/02/18 10:50:31]

I don't feel strongly about it: if y'all are happier with the current naming,
then keep it.

[msramek, 2016/02/18 12:48:43]

"superdomain" could be confusing, as in my mind this matches subdomains, it's
just that we implement it reversely. But I also don't have a strong opinion.

If Brett feels strongly, I guess I'll keep the naming for now :)

+
+  // A convenience method to produce an empty blacklist, a filter that matches
+  // everything.
+  static base::Callback<bool(const GURL&)> BuildEmptyFilter();
+
+ private:
+  // True if the origin of |url| is in the whitelist, or isn't in the blacklist.
+  bool MatchesURL(const GURL& url) const;
+
+  // True if any origin [scheme, host, port], such that |url| has the same
+  // scheme and port, and |url|'s host is the same or a subdomain of that host,
+  // is in the whitelist, or isn't in the blacklist.
+  bool MatchesURLWithSubdomains(const GURL& url) const;
+
+  // The list of origins and whether they should be interpreted as a whitelist
+  // or blacklist.
+  std::set<url::Origin> origin_list_;
+  Mode mode_;
+
+  // Whether the filter has already been built.
+  bool built_;
+
+  DISALLOW_COPY_AND_ASSIGN(OriginFilterBuilder);
+};
+
+#endif  // CHROME_BROWSER_BROWSING_DATA_ORIGIN_FILTER_BUILDER_H_