Remove CaseInsensitiveCompare from string_util.h

There were a number of callers in net using this for HTTP headers. I think
these callers actually just need ASCII case-insensitive comparisons so these
were changed.

The omnibox code used this functor. I added a new omnibox-specific one which
does not have the locale issues of the old string_util one, but which still
has the UTF-16 and combining accent issues (described in great detail in
the comment for this).

The Windows installer code can't depend on ICU so it calls the Win32 function
to do case-insensitive comparisons. This should match the system comparison
for registry keys better anyway.

I also changed a caller of StartsWith to use this version. I wrote this
StartsWith call using ToLower in a previous patch, but it turns out that the
lengths of case-mapped strings do change in practice, making the offset
computations of the suyrrounding code incorrect. This new version will be
like the old version (will miss some cases of case-insensitive equality) but
will handle 0x80-0xFF properly.

BUG=24917

Review URL: https://codereview.chromium.org/1230583014

Cr-Commit-Position: refs/heads/master@{#338624}

base/i18n/case_conversion.h

@@ -26,6 +26,10 @@ namespace i18n {
 // locale. Use this when comparing general Unicode strings that don't
 // necessarily belong in the user's current locale (like commands, protocol
 // names, other strings from the web) for case-insensitive equality.
+//
+// Note that case conversions will change the length of the string in some
+// not-uncommon cases. Never assume that the output is the same length as
+// the input.
 
 // Returns the lower case equivalent of string. Uses ICU's current locale.
 BASE_I18N_EXPORT string16 ToLower(StringPiece16 string);

base/strings/string_util.cc

@@ -101,6 +101,18 @@ template<> struct NonASCIIMask<8, wchar_t> {
 };
 #endif  // WCHAR_T_IS_UTF32
 
+// DO NOT USE. http://crbug.com/24917
+//
+// tolower() will given incorrect results for non-ASCII characters. Use the
+// ASCII version, base::i18n::ToLower, or base::i18n::FoldCase. This is here
+// for backwards-compat for StartsWith until such calls can be updated.
+struct CaseInsensitiveCompareDeprecated {
+ public:
+  bool operator()(char16 x, char16 y) const {
+    return tolower(x) == tolower(y);
+  }
+};
+
 }  // namespace
 
 namespace base {
@@ -611,7 +623,7 @@ bool StartsWith(const string16& str,
     if (search.size() > str.size())
       return false;
     return std::equal(search.begin(), search.end(), str.begin(),
-                      CaseInsensitiveCompare<char16>());
+                      CaseInsensitiveCompareDeprecated());
   }
   return StartsWith(StringPiece16(str), StringPiece16(search),
                     CompareCase::SENSITIVE);
@@ -667,10 +679,10 @@ bool EndsWith(const string16& str,
       return false;
     return std::equal(search.begin(), search.end(),
                       str.begin() + (str.size() - search.size()),
-                      CaseInsensitiveCompare<char16>());
+                      CaseInsensitiveCompareDeprecated());
   }
   return EndsWith(StringPiece16(str), StringPiece16(search),
-                    CompareCase::SENSITIVE);
+                  CompareCase::SENSITIVE);
 }
 
 char HexDigitToInt(wchar_t c) {

base/strings/string_util.h

@@ -103,20 +103,14 @@ template <class Char> inline Char ToUpperASCII(Char c) {
   return (c >= 'a' && c <= 'z') ? (c + ('A' - 'a')) : c;
 }
 
-// Function objects to aid in comparing/searching strings.
-
-// DO NOT USE. tolower() will given incorrect results for non-ASCII characters.
-// Use the ASCII version, base::i18n::ToLower, or base::i18n::FoldCase.
-template<typename Char> struct CaseInsensitiveCompare {
- public:
-  bool operator()(Char x, Char y) const {
-    // TODO(darin): Do we really want to do locale sensitive comparisons here?
-    // ANSWER(brettw): No.
-    // See http://crbug.com/24917
-    return tolower(x) == tolower(y);
-  }
-};
-
+// Functor for case-insensitive ASCII comparisons for STL algorithms like
+// std::search.
+//
+// Note that a full Unicode version of this functor is not possible to write
+// because case mappings might change the number of characters, depend on
+// context (combining accents), and require handling UTF-16. If you need
+// proper Unicode support, use base::i18n::ToLower/FoldCase and then just
+// use a normal operator== on the result.
 template<typename Char> struct CaseInsensitiveCompareASCII {
  public:
   bool operator()(Char x, Char y) const {

chrome/installer/util/shell_util.cc

@@ -746,8 +746,12 @@ class RegistryEntry {
       found = key.ReadValue(name_.c_str(), &read_value) == ERROR_SUCCESS;
       if (found) {
         correct_value = read_value.size() == value_.size() &&
-            std::equal(value_.begin(), value_.end(), read_value.begin(),
-                       base::CaseInsensitiveCompare<wchar_t>());
+            ::CompareString(LOCALE_USER_DEFAULT, NORM_IGNORECASE,
+                            read_value.data(),
+                            base::saturated_cast<int>(read_value.size()),
+                            value_.data(),
+                            base::saturated_cast<int>(value_.size()))
+                == CSTR_EQUAL;
       }
     } else {
       DWORD read_value;

components/omnibox/browser/autocomplete_i18n.h

@@ -0,0 +1,41 @@
+// Copyright 2015 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 COMPONENTS_OMNIBOX_BROWSER_AUTOCOMPLETE_I18N_H_
+#define COMPONENTS_OMNIBOX_BROWSER_AUTOCOMPLETE_I18N_H_
+
+#include "base/strings/string16.h"
+#include "third_party/icu/source/common/unicode/uchar.h"
+
+// Functor for a simple 16-bit Unicode case-insensitive comparison. This is
+// designed for the autocomplete system where we would rather get prefix lenths
+// correct than handle all possible case sensitivity issues.
+//
+// Any time this is used the result will be incorrect in some cases that
+// certain users will be able to discern. Ideally, this class would be deleted
+// and we would do full Unicode case-sensitivity mappings using
+// base::i18n::ToLower. However, ToLower can change the lenghts of strings,
+// making computations of offsets or prefix lengths difficult. Getting all
+// edge cases correct will require careful implementation and testing. In the
+// mean time, we use this simpler approach.
+//
+// This comparator will not handle combining accents properly since it compares
+// 16-bit values in isolation. If the two strings use the same sequence of
+// combining accents (this is the normal case) in both strings, it will work.
+//
+// Additionally, this comparator does not decode UTF sequences which is why it
+// is called "UCS2". UTF-16 surrogates will be compared literally (i.e. "case-
+// sensitively").
+//
+// There are also a few cases where the lower-case version of a character
+// expands to more than one code point that will not be handled properly. Such
+// characters will be compared case-sensitively.
+struct SimpleCaseInsensitiveCompareUCS2 {
+ public:
+  bool operator()(base::char16 x, base::char16 y) const {
+    return u_tolower(x) == u_tolower(y);
+  }
+};
+
+#endif  // COMPONENTS_OMNIBOX_BROWSER_AUTOCOMPLETE_I18N_H_

components/omnibox/browser/search_suggestion_parser.cc

@@ -15,6 +15,7 @@
 #include "base/strings/string_util.h"
 #include "base/strings/utf_string_conversions.h"
 #include "base/values.h"
+#include "components/omnibox/browser/autocomplete_i18n.h"
 #include "components/omnibox/browser/autocomplete_input.h"
 #include "components/omnibox/browser/url_prefix.h"
 #include "components/url_fixer/url_fixer.h"
@@ -158,7 +159,7 @@ void SearchSuggestionParser::SuggestResult::ClassifyMatchContents(
   // Do a case-insensitive search for |lookup_text|.
   base::string16::const_iterator lookup_position = std::search(
       match_contents_.begin(), match_contents_.end(), lookup_text.begin(),
-      lookup_text.end(), base::CaseInsensitiveCompare<base::char16>());
+      lookup_text.end(), SimpleCaseInsensitiveCompareUCS2());
   if (!allow_bolding_all && (lookup_position == match_contents_.end())) {
     // Bail if the code below to update the bolding would bold the whole
     // string.  Note that the string may already be entirely bolded; if

components/omnibox/browser/shortcuts_provider.cc

@@ -20,6 +20,7 @@
 #include "base/time/time.h"
 #include "components/history/core/browser/history_service.h"
 #include "components/metrics/proto/omnibox_input_type.pb.h"
+#include "components/omnibox/browser/autocomplete_i18n.h"
 #include "components/omnibox/browser/autocomplete_input.h"
 #include "components/omnibox/browser/autocomplete_match.h"
 #include "components/omnibox/browser/autocomplete_provider_client.h"
@@ -218,9 +219,11 @@ AutocompleteMatch ShortcutsProvider::ShortcutToACMatch(
   // input of "foo.c" to autocomplete to "foo.com" for a fill_into_edit of
   // "http://foo.com".
   if (AutocompleteMatch::IsSearchType(match.type)) {
-    if (base::StartsWith(base::i18n::ToLower(match.fill_into_edit),
-                         base::i18n::ToLower(input.text()),
-                         base::CompareCase::SENSITIVE)) {
+    if (match.fill_into_edit.size() >= input.text().size() &&
+        std::equal(match.fill_into_edit.begin(),
+                   match.fill_into_edit.begin() + input.text().size(),
+                   input.text().begin(),
+                   SimpleCaseInsensitiveCompareUCS2())) {
       match.inline_autocompletion =
           match.fill_into_edit.substr(input.text().length());
       match.allowed_to_be_default_match =

extensions/browser/api/declarative_webrequest/webrequest_action.cc

@@ -970,13 +970,8 @@ WebRequestRemoveResponseHeaderAction::CreateDelta(
   void* iter = NULL;
   std::string current_value;
   while (headers->EnumerateHeader(&iter, name_, &current_value)) {
-    if (has_value_ &&
-           (current_value.size() != value_.size() ||
-            !std::equal(current_value.begin(), current_value.end(),
-                        value_.begin(),
-                        base::CaseInsensitiveCompare<char>()))) {
+    if (has_value_ && !base::EqualsCaseInsensitiveASCII(current_value, value_))
       continue;
-    }
     result->deleted_response_headers.push_back(make_pair(name_, current_value));
   }
   return result;

net/http/http_response_headers.cc

@@ -603,9 +603,7 @@ bool HttpResponseHeaders::HasHeaderValue(const base::StringPiece& name,
   void* iter = NULL;
   std::string temp;
   while (EnumerateHeader(&iter, name, &temp)) {
-    if (value.size() == temp.size() &&
-        std::equal(temp.begin(), temp.end(), value.begin(),
-                   base::CaseInsensitiveCompare<char>()))
+    if (base::EqualsCaseInsensitiveASCII(value, temp))
       return true;
   }
   return false;
@@ -743,11 +741,8 @@ size_t HttpResponseHeaders::FindHeader(size_t from,
   for (size_t i = from; i < parsed_.size(); ++i) {
     if (parsed_[i].is_continuation())
       continue;
-    const std::string::const_iterator& name_begin = parsed_[i].name_begin;
-    const std::string::const_iterator& name_end = parsed_[i].name_end;
-    if (static_cast<size_t>(name_end - name_begin) == search.size() &&
-        std::equal(name_begin, name_end, search.begin(),
-                   base::CaseInsensitiveCompare<char>()))
+    base::StringPiece name(parsed_[i].name_begin, parsed_[i].name_end);
+    if (base::EqualsCaseInsensitiveASCII(search, name))
       return i;
   }
 

net/url_request/url_request_unittest.cc

@@ -258,12 +258,9 @@ void TestLoadTimingNoHttpResponse(
 
 // Do a case-insensitive search through |haystack| for |needle|.
 bool ContainsString(const std::string& haystack, const char* needle) {
-  std::string::const_iterator it =
-      std::search(haystack.begin(),
-                  haystack.end(),
-                  needle,
-                  needle + strlen(needle),
-                  base::CaseInsensitiveCompare<char>());
+  std::string::const_iterator it = std::search(
+      haystack.begin(), haystack.end(), needle, needle + strlen(needle),
+      base::CaseInsensitiveCompareASCII<char>());
   return it != haystack.end();
 }
 

win8/test/ui_automation_client.cc

@@ -323,10 +323,11 @@ void UIAutomationClient::Context::HandleWindowOpen(
 
   base::string16 class_name(V_BSTR(var.ptr()));
 
-  // Window class names are atoms, which are case-insensitive.
+  // Window class names are atoms, which are case-insensitive. Assume that
+  // the window in question only needs ASCII case-insensitivity.
   if (class_name.size() == class_name_.size() &&
       std::equal(class_name.begin(), class_name.end(), class_name_.begin(),
-                 base::CaseInsensitiveCompare<wchar_t>())) {
+                 base::CaseInsensitiveCompareASCII<wchar_t>())) {
     RemoveWindowObserver();
     ProcessWindow(window);
   }