Use ICU case conversion/transliterator for case conversion behind a flag

src/runtime/runtime-i18n.cc

 // Copyright 2014 the V8 project authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 
 #ifdef V8_I18N_SUPPORT
 #include "src/runtime/runtime-utils.h"
 
 #include "src/api.h"
 #include "src/api-natives.h"
@@ skipping 11 matching lines @@
 #include "unicode/dcfmtsym.h"
 #include "unicode/decimfmt.h"
 #include "unicode/dtfmtsym.h"
 #include "unicode/dtptngen.h"
 #include "unicode/locid.h"
 #include "unicode/numfmt.h"
 #include "unicode/numsys.h"
 #include "unicode/rbbi.h"
 #include "unicode/smpdtfmt.h"
 #include "unicode/timezone.h"
+#include "unicode/translit.h"
 #include "unicode/uchar.h"
 #include "unicode/ucol.h"
 #include "unicode/ucurr.h"
 #include "unicode/uloc.h"
+#include "unicode/unistr.h"
 #include "unicode/unum.h"
 #include "unicode/uversion.h"
 
 
 namespace v8 {
 namespace internal {
 
 RUNTIME_FUNCTION(Runtime_CanonicalizeLanguageTag) {
   HandleScope scope(isolate);
   Factory* factory = isolate->factory();
@@ skipping 696 matching lines @@
   } else if (status >= UBRK_WORD_LETTER && status < UBRK_WORD_LETTER_LIMIT) {
     return *isolate->factory()->NewStringFromStaticChars("letter");
   } else if (status >= UBRK_WORD_KANA && status < UBRK_WORD_KANA_LIMIT) {
     return *isolate->factory()->NewStringFromStaticChars("kana");
   } else if (status >= UBRK_WORD_IDEO && status < UBRK_WORD_IDEO_LIMIT) {
     return *isolate->factory()->NewStringFromStaticChars("ideo");
   } else {
     return *isolate->factory()->NewStringFromStaticChars("unknown");
   }
 }
+
+namespace {
+inline void LocaleConvertCaseHelper(icu::UnicodeString* s, bool is_to_upper,
+                                    const icu::Locale& locale) {
+  // TODO(jshin): Greek and Lithuanian need to use 'transliterator'.
+  if (is_to_upper)
+    s->toUpper(locale);
+  else
+    s->toLower(locale);
+}
+void ConvertCaseWithTransliterator(icu::UnicodeString* input, bool is_to_upper,
+                                   const char* language) {
+#if 0
+  char transliteratorId[8];
+
+  strncpy(transliteratorId, language, 2);
+  transliteratorId[2] = '-';
+  strncpy(transliteratorId + 3, is_to_upper ? "Lower" : "Upper", 5);
+#endif
+
+  icu::UnicodeString transliteratorId;
+  if (strcmp("el", language) == 0) {
+    transliteratorId = UNICODE_STRING_SIMPLE("el-Upper");
+  } else if (strcmp("lt", language) == 0) {
+    transliteratorId = is_to_upper ? UNICODE_STRING_SIMPLE("lt-Upper")
+                                   : UNICODE_STRING_SIMPLE("lt-Lower");
+  }
+
+  UErrorCode status = U_ZERO_ERROR;
+  base::SmartPointer<icu::Transliterator> translit(
+      icu::Transliterator::createInstance(transliteratorId, UTRANS_FORWARD,
+                                          status));
+  // RUNTIME_ASSERT(U_SUCCESS(status));
+  if (U_FAILURE(status)) return;
+  translit->transliterate(*input);
+}
+
+MUST_USE_RESULT Object* LocaleConvertCase(Handle<String> s, Isolate* isolate,
+                                          bool is_to_upper, int locale_id) {
+  static const char* case_conversion_languages[] = {
+      "az", "el", "lt", "tr",
+  };
+  RUNTIME_ASSERT(locale_id >= -1 &&
+                 locale_id <
+                     static_cast<int>(arraysize(case_conversion_languages)));
+  int32_t length = s->length();
+  icu::UnicodeString converted;
+  {
+    DisallowHeapAllocation no_gc;
+    DCHECK(s->IsFlat());
+    String::FlatContent flat = s->GetFlatContent();
+
+    const UChar* src;
+    if (flat.IsOneByte()) {
+      base::SmartArrayPointer<uc16> sap = s->ToWideCString();
+      src = reinterpret_cast<const UChar*>(sap.get());
+      converted = icu::UnicodeString(src, length);
+    } else {
+      src = reinterpret_cast<const UChar*>(flat.ToUC16Vector().start());
+      converted = icu::UnicodeString(src, length);
+    }
+  }
+#if 0
+  v8::String::Value string_value(v8::Utils::ToLocal(s));
+  icu::UnicodeString converted(
+      false, reinterpret_cast<const UChar*>(*string_value), length);

[jungshik at Google, 2016/04/11 17:47:29]

a question: Using this public API is more concise than using the internal
mechanism. I just need to extract 'UChar string' (buffer and length). Is there a
perf (and other reason) to use a more verbose internal way (the code above this
commented-out block)?  BTW, 'normalize' method (in runtime-i18n.cc) is
implemented with String::Value API. 

+#endif
+
+  if (locale_id == -1) {
+    LocaleConvertCaseHelper(&converted, is_to_upper, icu::Locale::getRoot());
+  } else {
+    const char* case_conversion_language = case_conversion_languages[locale_id];
+    // el-Upper, lt-Upper, lt-Lower need to be done with transliterator.
+    if ((locale_id == 1 && is_to_upper) || locale_id == 2) {
+      ConvertCaseWithTransliterator(&converted, is_to_upper,
+                                    case_conversion_language);
+    } else {
+      LocaleConvertCaseHelper(&converted, is_to_upper,
+                              icu::Locale(case_conversion_language));
+    }
+  }
+
+  Handle<String> result;
+  ASSIGN_RETURN_FAILURE_ON_EXCEPTION(
+      isolate, result,
+      isolate->factory()->NewStringFromTwoByte(Vector<const uint16_t>(
+          reinterpret_cast<const uint16_t*>(converted.getBuffer()),
+          converted.length())));
+  return *result;
+}
+
+inline bool IsASCIIUpper(uint16_t ch) { return ch >= 'A' && ch <= 'Z'; }
+
+inline uint16_t ToASCIILower(uint16_t ch) {
+  return ch | ((ch >= 'A' && ch <= 'Z') << 5);
+}
+
+inline uint16_t ToASCIIUpper(uint16_t ch) {
+  return ch & ~((ch >= 'a' && ch <= 'z') << 5);
+}
+
+}  // namespace
+
+RUNTIME_FUNCTION(Runtime_StringToLowerCaseI18N) {
+  HandleScope scope(isolate);
+  DCHECK_EQ(args.length(), 1);
+  CONVERT_ARG_HANDLE_CHECKED(String, s, 0);
+
+  int length = s->length();
+  s = String::Flatten(s);
+  // First scan the string for uppercase and non-ASCII characters:
+  if (s->HasOnlyOneByteChars()) {
+    unsigned first_index_to_lower = length;
+    for (int index = 0; index < length; ++index) {
+      // Blink specializes this path for one-byte strings, so it
+      // does not need to do a generic get, but can do the equivalent
+      // of SeqOneByteStringGet.
+      uint16_t ch = s->Get(index);
+      if (V8_UNLIKELY(IsASCIIUpper(ch) || ch & ~0x7F)) {
+        first_index_to_lower = index;
+        break;
+      }
+    }
+
+    // Nothing to do if the string is all ASCII with no uppercase.
+    if (first_index_to_lower == length) return *s;
+
+    // We depend here on the invariant that the length of a Latin1
+    // string is invariant under ToLowerCase, and the result always
+    // fits in the Latin1 range (untrue for ToUpperCase, and might
+    // be untrue in some locales, but this is the root locale)
+    Handle<SeqOneByteString> result;
+    ASSIGN_RETURN_FAILURE_ON_EXCEPTION(
+        isolate, result, isolate->factory()->NewRawOneByteString(length));
+    if (s->IsSeqOneByteString()) {
+      SeqOneByteString* source = SeqOneByteString::cast(*s);
+      CopyChars(result->GetChars(), source->GetChars(), first_index_to_lower);
+    } else {
+      // Do we have to worry about External{One,Two}ByteString?
+      DCHECK(s->IsSeqTwoByteString());
+      SeqTwoByteString* source = SeqTwoByteString::cast(*s);
+      CopyChars(result->GetChars(), source->GetChars(), first_index_to_lower);
+    }
+
+    for (int index = first_index_to_lower; index < length; ++index) {
+      uint16_t ch = s->Get(index);
+      result->SeqOneByteStringSet(
+          index, V8_UNLIKELY(ch & ~0x7F) ? static_cast<uint16_t>(u_tolower(ch))
+                                         : ToASCIILower(ch));
+    }
+
+    return *result;
+  }
+
+  // Blink had an additional case here for ASCII 2-byte strings, but
+  // that is subsumed by the above code (assuming there isn't a false
+  // negative for HasOnlyOneByteChars).
+
+  // Do a slower implementation for cases that include non-ASCII characters.
+  return LocaleConvertCase(s, isolate, false, -1);
+}
+
+RUNTIME_FUNCTION(Runtime_StringToUpperCaseI18N) {
+  HandleScope scope(isolate);
+  DCHECK_EQ(args.length(), 1);
+  CONVERT_ARG_HANDLE_CHECKED(String, s, 0);
+
+  // This function could be optimized for no-op cases the way lowercase
+  // counterpart is, but in empirical testing, few actual calls to upper()
+  // are no-ops. So, it wouldn't be worth the extra time for pre-scanning.
+
+  int32_t length = s->length();
+  s = String::Flatten(s);
+  static const uint16_t sharp_s = 0x00DFu;
+
+  if (s->HasOnlyOneByteChars()) {
+    Handle<SeqOneByteString> result;
+    ASSIGN_RETURN_FAILURE_ON_EXCEPTION(
+        isolate, result, isolate->factory()->NewRawOneByteString(length));
+
+    // Do a faster loop for the case where all the characters are ASCII.
+    uint16_t ored = 0;
+    for (int index = 0; index < length; ++index) {
+      uint16_t ch = s->Get(index);
+      ored |= ch;
+      result->SeqOneByteStringSet(index, ToASCIIUpper(ch));
+    }
+    if (!(ored & ~0x7F)) return *result;
+
+    // Do a slower implementation for cases that include non-ASCII Latin-1
+    // characters.
+    int sharp_s_count = 0;
+
+    // There are two special cases.
+    //  1. latin-1 characters when converted to upper case are 16 bit
+    //  characters.
+    //  2. Lower case sharp-S converts to "SS" (two characters)
+    for (int32_t index = 0; index < length; ++index) {
+      uint16_t ch = s->Get(index);
+      if (V8_UNLIKELY(ch == sharp_s)) {
+        ++sharp_s_count;
+        continue;
+      }
+      uint16_t upper = static_cast<uint16_t>(u_toupper(static_cast<UChar>(ch)));
+      if (V8_UNLIKELY(upper > 0xff)) {
+        // Since this upper-cased character does not fit in an 8-bit string, we
+        // need to take the 16-bit path.
+        return LocaleConvertCase(s, isolate, true, -1);
+      }
+      result->SeqOneByteStringSet(index, upper);
+    }
+
+    if (sharp_s_count == 0) return *result;
+
+    // We have sharp_s_count sharp-s characters, but none of the other special
+    // characters.)
+    ASSIGN_RETURN_FAILURE_ON_EXCEPTION(
+        isolate, result,
+        isolate->factory()->NewRawOneByteString(length + sharp_s_count));
+    for (int32_t index = 0, dest_index = 0; index < length; ++index) {
+      uint16_t ch = s->Get(index);
+      if (ch == sharp_s) {
+        result->SeqOneByteStringSet(dest_index++, 'S');
+        result->SeqOneByteStringSet(dest_index++, 'S');
+      } else {
+        uint16_t upper =
+            static_cast<uint16_t>(u_toupper(static_cast<UChar>(ch)));
+        result->SeqOneByteStringSet(dest_index++, upper);
+      }
+    }
+
+    return *result;
+  }
+
+  return LocaleConvertCase(s, isolate, true, -1);
+}
+
+RUNTIME_FUNCTION(Runtime_StringLocaleConvertCase) {
+  HandleScope scope(isolate);
+  DCHECK_EQ(args.length(), 3);
+  CONVERT_ARG_HANDLE_CHECKED(String, s, 0);
+  CONVERT_BOOLEAN_ARG_CHECKED(is_upper, 1);
+  CONVERT_NUMBER_CHECKED(int, lang_id, Int32, args[2]);
+
+  return LocaleConvertCase(s, isolate, is_upper, lang_id);
+}
+
 }  // namespace internal
 }  // namespace v8
 
 #endif  // V8_I18N_SUPPORT