Reduce the perf overhead of our dangling markup experiments.

url/url_util.cc

 // Copyright 2013 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 "url/url_util.h"
 
 #include <stddef.h>
 #include <string.h>
 
 #include "base/debug/leak_annotations.h"
@@ skipping 181 matching lines @@
                     CharsetConverter* charset_converter,
                     CanonOutput* output,
                     Parsed* output_parsed) {
   // Reserve enough room in the output for the input, plus some extra so that
   // we have room if we have to escape a few things without reallocating.
   output->ReserveSizeIfNeeded(spec_len + 8);
 
   // Remove any whitespace from the middle of the relative URL if necessary.
   // Possibly this will result in copying to the new buffer.
   RawCanonOutputT<CHAR> whitespace_buffer;
-  if (whitespace_policy == REMOVE_WHITESPACE)
-    spec = RemoveURLWhitespace(spec, spec_len, &whitespace_buffer, &spec_len);
+  if (whitespace_policy == REMOVE_WHITESPACE) {
+    int original_len = spec_len;
+    spec =
+        RemoveURLWhitespace(spec, original_len, &whitespace_buffer, &spec_len);
+    if (spec_len != original_len)
+      output_parsed->whitespace_removed = true;
+  }
 
   Parsed parsed_input;
 #ifdef WIN32
   // For Windows, we allow things that look like absolute Windows paths to be
   // fixed up magically to file URLs. This is done for IE compatibility. For
   // example, this will change "c:/foo" into a file URL rather than treating
   // it as a URL with the protocol "c". It also works for UNC ("\\foo\bar.txt").
   // There is similar logic in url_canon_relative.cc for
   //
   // For Max & Unix, we don't do this (the equivalent would be "/foo/bar" which
@@ skipping 59 matching lines @@
                        CharsetConverter* charset_converter,
                        CanonOutput* output,
                        Parsed* output_parsed) {
   // Remove any whitespace from the middle of the relative URL, possibly
   // copying to the new buffer.
   RawCanonOutputT<CHAR> whitespace_buffer;
   int relative_length;
   const CHAR* relative = RemoveURLWhitespace(in_relative, in_relative_length,
                                              &whitespace_buffer,
                                              &relative_length);
+  if (in_relative_length != relative_length)
+    output_parsed->whitespace_removed = true;
+
   bool base_is_authority_based = false;
   bool base_is_hierarchical = false;
   if (base_spec &&
       base_parsed.scheme.is_nonempty()) {
     int after_scheme = base_parsed.scheme.end() + 1;  // Skip past the colon.
     int num_slashes = CountConsecutiveSlashes(base_spec, after_scheme,
                                               base_spec_len);
     base_is_authority_based = num_slashes > 1;
     base_is_hierarchical = num_slashes > 0;
   }
@@ skipping 500 matching lines @@
   return DoCompareSchemeComponent(spec, component, compare_to);
 }
 
 bool CompareSchemeComponent(const base::char16* spec,
                             const Component& component,
                             const char* compare_to) {
   return DoCompareSchemeComponent(spec, component, compare_to);
 }
 
 }  // namespace url