Reduce buggy usage of the registry controlled domain service.

url/url_canon.h

Index: url/url_canon.h
diff --git a/url/url_canon.h b/url/url_canon.h
index 95d53453f64362efc98d8141c11691573eaae165..2fa145337081f4796d0d8a87bb4f81b67977847b 100644
--- a/url/url_canon.h
+++ b/url/url_canon.h
@@ -379,6 +379,31 @@ URL_EXPORT void CanonicalizeHostVerbose(const base::char16* spec,
                                         CanonOutput* output,
                                         CanonHostInfo* host_info);
 
+// Canonicalizes a string according to the host canonicalization rules. Unlike
+// CanonicalizeHost, this will not check for IP addresses which can change the
+// meaning (and canonicalization) of the components. This means it is possible
+// to call this for sub-components of a host name without corruption.
+//
+// As an example, "01.02.03.04.com" is a canonical hostname. If you called
+// CanonicalizeHost on the substring "01.02.03.04" it will get "fixed" to
+// "1.2.3.4" which will produce an invalid host name when reassembled. This
+// is more common because all numbers by themselves are considered IP

[Peter Kasting, 2016/10/22 05:04:20]

Nit: "more common" than what?  More common than one might think reading this
comment?

[Peter Kasting, 2016/10/25 01:33:32]

(Still confused about this)

[brettw, 2016/10/25 20:28:17]

Done.

+// addresses, so "5" canonicalizes to "0.0.0.5".
+//
+// Be careful, unless the input is guaranteed ASCII, it's not possible to split

[Peter Kasting, 2016/10/22 05:04:20]

Nit: first comma -> semicolon

+// host names on anything but "." because Punycode works on each dot-separated
+// substring as a unit.

[Peter Kasting, 2016/10/22 05:04:20]

Nit: Maybe this paragraph should be more direct: "Because Punycode works on each
dot-separated substring as a unit, you should only pass this function substrings
that represent complete dot-separated subcomponents of the original host."

(Do we really need to put in an ASCII caveat?)

[brettw, 2016/10/24 21:45:24]

I mostly did your comment with some addition.

+//
+// Returns true if the host was valid. This function will treat a 0-length
+// host as valid (because it's designed to be used for substrings) while the
+// full version above will mark empty hosts as broken.
+URL_EXPORT bool CanonicalizeHostSubstring(const char* spec,
+                                          const Component& host,
+                                          CanonOutput* output);
+URL_EXPORT bool CanonicalizeHostSubstring(const base::char16* spec,
+                                          const Component& host,
+                                          CanonOutput* output);
+
 // IP addresses.
 //
 // Tries to interpret the given host name as an IPv4 or IPv6 address. If it is