Teach SiteInstance::GetSiteForURL() about blob and filesystem URLs.

chrome/browser/search/search.cc

 // Copyright 2012 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.
 
 #include "chrome/browser/search/search.h"
 
 #include <stddef.h>
 
 #include "base/command_line.h"
 #include "base/metrics/field_trial.h"
@@ skipping 475 matching lines @@
       << "Error granting Instant access.";
 
   if (url.SchemeIs(chrome::kChromeSearchScheme))
     return url;
 
   GURL effective_url(url);
 
   // Replace the scheme with "chrome-search:".
   url::Replacements<char> replacements;
   std::string search_scheme(chrome::kChromeSearchScheme);
+  replacements.ClearPort();

[Charlie Reis, 2016/04/22 18:11:09]

I'll defer to Samarth or someone familiar with the NTP code here, since this
would be a problem if we later pulled the host out and tried to use it.  It does
seem a little counter-intuitive to me, since the only reason we need to do this
is because the scheme was declared as not having a port.

I'd say that we should declare the scheme as having a port instead, since it
does have a network host embedded in it in some (if not all) cases.

[ncarter (slow), 2016/04/22 20:00:34]

We're erasing the underlying scheme anyway (which is usually https://, except in
tests, where we allow it to be overridden) so even if we preserve the port,
conversion from a chrome-search:// URL back to a network URL doesn't seem
possible without extra information.

I'm guessing that's why this extra erasure doesn't break tests -- they'd already
be broken if they assumed https?

Or to think about it differently: if we think of the chrome-search scheme as
having a network port, is the default value 80 or 443?

[Charlie Reis, 2016/04/22 20:20:46]

That's a good point about losing the scheme.  That does suggest that the port is
unlikely to be needed from a chrome-search:// URL, since we can't meaningfully
recover the original URL from this one.

Thus, I think I'm ok with leaving it as SCHEME_WITHOUT_PORT and clearing it
here.  (Please add a comment about it, though, so we remember if we come back to
it later.)

The downside is that the chrome-search://foo.com URL can't distinguish between
variations of foo.com if the default search provider changes (e.g., between
ports or http/https), but that's of course a problem already today with
http/https.

   replacements.SetScheme(search_scheme.data(),
                          url::Component(0, search_scheme.length()));
 
   // If this is the URL for a server-provided NTP, replace the host with
   // "remote-ntp".
   std::string remote_ntp_host(chrome::kChromeSearchRemoteNtpHost);
   NewTabURLDetails details = NewTabURLDetails::ForProfile(profile);
   if (details.state == NEW_TAB_URL_VALID &&
       MatchesOriginAndPath(url, details.url)) {
     replacements.SetHost(remote_ntp_host.c_str(),
@@ skipping 78 matching lines @@
       kUseAltInstantURL, false, flags);
 }
 
 bool ShouldUseSearchPathForInstant() {
   FieldTrialFlags flags;
   return GetFieldTrialInfo(&flags) && GetBoolValueForFlagWithDefault(
       kUseSearchPathForInstant, false, flags);
 }
 
 }  // namespace search