OLD | NEW |
---|---|
1 // Copyright 2014 the V8 project authors. All rights reserved. | 1 // Copyright 2014 the V8 project authors. All rights reserved. |
2 // Use of this source code is governed by a BSD-style license that can be | 2 // Use of this source code is governed by a BSD-style license that can be |
3 // found in the LICENSE file. | 3 // found in the LICENSE file. |
4 | 4 |
5 | 5 |
6 #ifdef V8_I18N_SUPPORT | 6 #ifdef V8_I18N_SUPPORT |
7 #include "src/runtime/runtime-utils.h" | 7 #include "src/runtime/runtime-utils.h" |
8 | 8 |
9 #include <memory> | 9 #include <memory> |
10 | 10 |
11 #include "src/api-natives.h" | |
11 #include "src/api.h" | 12 #include "src/api.h" |
12 #include "src/api-natives.h" | |
13 #include "src/arguments.h" | 13 #include "src/arguments.h" |
14 #include "src/factory.h" | 14 #include "src/factory.h" |
15 #include "src/i18n.h" | 15 #include "src/i18n.h" |
16 #include "src/isolate-inl.h" | 16 #include "src/isolate-inl.h" |
17 #include "src/messages.h" | 17 #include "src/messages.h" |
18 #include "src/utils.h" | |
18 | 19 |
19 #include "unicode/brkiter.h" | 20 #include "unicode/brkiter.h" |
20 #include "unicode/calendar.h" | 21 #include "unicode/calendar.h" |
21 #include "unicode/coll.h" | 22 #include "unicode/coll.h" |
22 #include "unicode/curramt.h" | 23 #include "unicode/curramt.h" |
23 #include "unicode/datefmt.h" | 24 #include "unicode/datefmt.h" |
24 #include "unicode/dcfmtsym.h" | 25 #include "unicode/dcfmtsym.h" |
25 #include "unicode/decimfmt.h" | 26 #include "unicode/decimfmt.h" |
26 #include "unicode/dtfmtsym.h" | 27 #include "unicode/dtfmtsym.h" |
27 #include "unicode/dtptngen.h" | 28 #include "unicode/dtptngen.h" |
(...skipping 1056 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... | |
1084 | 1085 |
1085 } // namespace | 1086 } // namespace |
1086 | 1087 |
1087 RUNTIME_FUNCTION(Runtime_StringToLowerCaseI18N) { | 1088 RUNTIME_FUNCTION(Runtime_StringToLowerCaseI18N) { |
1088 HandleScope scope(isolate); | 1089 HandleScope scope(isolate); |
1089 DCHECK_EQ(args.length(), 1); | 1090 DCHECK_EQ(args.length(), 1); |
1090 CONVERT_ARG_HANDLE_CHECKED(String, s, 0); | 1091 CONVERT_ARG_HANDLE_CHECKED(String, s, 0); |
1091 | 1092 |
1092 int length = s->length(); | 1093 int length = s->length(); |
1093 s = String::Flatten(s); | 1094 s = String::Flatten(s); |
1094 // First scan the string for uppercase and non-ASCII characters: | 1095 |
1096 bool is_ascii = true; | |
1095 if (s->HasOnlyOneByteChars()) { | 1097 if (s->HasOnlyOneByteChars()) { |
1096 int first_index_to_lower = length; | 1098 // Scan the string for uppercase and non-ASCII characters for strings |
1097 for (int index = 0; index < length; ++index) { | 1099 // shorter than a machine-word without any memory allocation overhead. |
Yang
2016/12/05 19:19:29
What is the rationale for doing this only to short
| |
1098 // Blink specializes this path for one-byte strings, so it | 1100 int index_to_first_upper = length; |
1099 // does not need to do a generic get, but can do the equivalent | 1101 if (static_cast<size_t>(length) < sizeof(uintptr_t)) { |
1100 // of SeqOneByteStringGet. | 1102 for (int index = 0; index < length; ++index) { |
1101 uint16_t ch = s->Get(index); | 1103 uint16_t ch = s->Get(index); |
1102 if (V8_UNLIKELY(IsASCIIUpper(ch) || ch & ~0x7F)) { | 1104 if (V8_UNLIKELY(IsASCIIUpper(ch) || ch & ~0x7F)) { |
1103 first_index_to_lower = index; | 1105 is_ascii = !(ch & ~0x7F); |
1104 break; | 1106 index_to_first_upper = index; |
1107 break; | |
1108 } | |
1105 } | 1109 } |
1110 // Nothing to do if the string is all ASCII with no uppercase. | |
1111 if (index_to_first_upper == length) return *s; | |
1106 } | 1112 } |
Yang
2016/12/05 19:19:29
So if the string is longer than a word, we always
| |
1107 | 1113 |
1108 // Nothing to do if the string is all ASCII with no uppercase. | |
1109 if (first_index_to_lower == length) return *s; | |
1110 | |
1111 // We depend here on the invariant that the length of a Latin1 | 1114 // We depend here on the invariant that the length of a Latin1 |
1112 // string is invariant under ToLowerCase, and the result always | 1115 // string is invariant under ToLowerCase, and the result always |
1113 // fits in the Latin1 range in the *root locale*. It does not hold | 1116 // fits in the Latin1 range in the *root locale*. It does not hold |
1114 // for ToUpperCase even in the root locale. | 1117 // for ToUpperCase even in the root locale. |
1115 Handle<SeqOneByteString> result; | 1118 Handle<SeqOneByteString> result; |
1116 ASSIGN_RETURN_FAILURE_ON_EXCEPTION( | 1119 ASSIGN_RETURN_FAILURE_ON_EXCEPTION( |
1117 isolate, result, isolate->factory()->NewRawOneByteString(length)); | 1120 isolate, result, isolate->factory()->NewRawOneByteString(length)); |
1118 | 1121 |
1119 DisallowHeapAllocation no_gc; | 1122 DisallowHeapAllocation no_gc; |
1120 String::FlatContent flat = s->GetFlatContent(); | 1123 String::FlatContent flat = s->GetFlatContent(); |
1124 uint8_t* dest = result->GetChars(); | |
1125 // Instead of checking is_ascii here, we'd better modify FastAsciiConvert | |
Yang
2016/12/05 19:19:29
Is this a TODO?
| |
1126 // to return the index to the first non-ASCII character. | |
1127 if (flat.IsOneByte() && is_ascii) { | |
1128 const uint8_t* src = flat.ToOneByteVector().start(); | |
1129 bool has_changed_character = false; | |
1130 bool is_ascii = FastAsciiConvert<true>(reinterpret_cast<char*>(dest), | |
1131 reinterpret_cast<const char*>(src), | |
1132 length, &has_changed_character); | |
1133 // If not ASCII, we discard the result and start anew. | |
1134 if (is_ascii) return has_changed_character ? *result : *s; | |
1135 } | |
1136 | |
1137 if (index_to_first_upper == length) { | |
1138 for (int index = 0; index < length; ++index) { | |
1139 uint16_t ch = s->Get(index); | |
1140 if (V8_UNLIKELY(IsASCIIUpper(ch) || ch & ~0x7F)) { | |
1141 index_to_first_upper = index; | |
1142 break; | |
1143 } | |
1144 } | |
1145 } | |
1146 | |
1121 if (flat.IsOneByte()) { | 1147 if (flat.IsOneByte()) { |
1148 // An ASCII input without any uppercase characters is already handled by | |
1149 // the short-string scanner and FastAsciiConvert. | |
1150 DCHECK(index_to_first_upper < length); | |
1122 const uint8_t* src = flat.ToOneByteVector().start(); | 1151 const uint8_t* src = flat.ToOneByteVector().start(); |
1123 CopyChars(result->GetChars(), src, | 1152 CopyChars(dest, src, static_cast<size_t>(index_to_first_upper)); |
1124 static_cast<size_t>(first_index_to_lower)); | 1153 for (int index = index_to_first_upper; index < length; ++index) { |
1125 for (int index = first_index_to_lower; index < length; ++index) { | |
1126 uint16_t ch = static_cast<uint16_t>(src[index]); | 1154 uint16_t ch = static_cast<uint16_t>(src[index]); |
1127 result->SeqOneByteStringSet(index, ToLatin1Lower(ch)); | 1155 result->SeqOneByteStringSet(index, ToLatin1Lower(ch)); |
1128 } | 1156 } |
1129 } else { | 1157 } else { |
1158 // Nothing to do if the string is all ASCII with no uppercase. | |
1159 if (index_to_first_upper == length) return *s; | |
1130 const uint16_t* src = flat.ToUC16Vector().start(); | 1160 const uint16_t* src = flat.ToUC16Vector().start(); |
1131 CopyChars(result->GetChars(), src, | 1161 CopyChars(dest, src, static_cast<size_t>(index_to_first_upper)); |
1132 static_cast<size_t>(first_index_to_lower)); | 1162 for (int index = index_to_first_upper; index < length; ++index) { |
1133 for (int index = first_index_to_lower; index < length; ++index) { | |
1134 uint16_t ch = src[index]; | 1163 uint16_t ch = src[index]; |
1135 result->SeqOneByteStringSet(index, ToLatin1Lower(ch)); | 1164 result->SeqOneByteStringSet(index, ToLatin1Lower(ch)); |
1136 } | 1165 } |
1137 } | 1166 } |
1138 | 1167 |
1139 return *result; | 1168 return *result; |
1140 } | 1169 } |
1141 | 1170 |
1142 // Blink had an additional case here for ASCII 2-byte strings, but | 1171 // Blink had an additional case here for ASCII 2-byte strings, but |
1143 // that is subsumed by the above code (assuming there isn't a false | 1172 // that is subsumed by the above code (assuming there isn't a false |
1144 // negative for HasOnlyOneByteChars). | 1173 // negative for HasOnlyOneByteChars). |
1145 | 1174 |
1146 // Do a slower implementation for cases that include non-ASCII characters. | 1175 // Do a slower implementation for cases that include non-ASCII characters. |
1147 return LocaleConvertCase(s, isolate, false, ""); | 1176 return LocaleConvertCase(s, isolate, false, ""); |
1148 } | 1177 } |
1149 | 1178 |
1150 RUNTIME_FUNCTION(Runtime_StringToUpperCaseI18N) { | 1179 RUNTIME_FUNCTION(Runtime_StringToUpperCaseI18N) { |
1151 HandleScope scope(isolate); | 1180 HandleScope scope(isolate); |
1152 DCHECK_EQ(args.length(), 1); | 1181 DCHECK_EQ(args.length(), 1); |
1153 CONVERT_ARG_HANDLE_CHECKED(String, s, 0); | 1182 CONVERT_ARG_HANDLE_CHECKED(String, s, 0); |
1154 | 1183 |
1155 // This function could be optimized for no-op cases the way lowercase | |
1156 // counterpart is, but in empirical testing, few actual calls to upper() | |
1157 // are no-ops. So, it wouldn't be worth the extra time for pre-scanning. | |
1158 | |
1159 int32_t length = s->length(); | 1184 int32_t length = s->length(); |
1160 s = String::Flatten(s); | 1185 s = String::Flatten(s); |
1161 | 1186 |
1162 if (s->HasOnlyOneByteChars()) { | 1187 if (s->HasOnlyOneByteChars()) { |
1188 #if 0 | |
1163 Handle<SeqOneByteString> result; | 1189 Handle<SeqOneByteString> result; |
1164 ASSIGN_RETURN_FAILURE_ON_EXCEPTION( | 1190 ASSIGN_RETURN_FAILURE_ON_EXCEPTION( |
1165 isolate, result, isolate->factory()->NewRawOneByteString(length)); | 1191 isolate, result, isolate->factory()->NewRawOneByteString(length)); |
1192 #endif | |
1193 Handle<SeqOneByteString> result = | |
1194 isolate->factory()->NewRawOneByteString(length).ToHandleChecked(); | |
jungshik at Google
2016/12/02 06:52:38
The first part of Runtime_STringToUpperCaseI18N fo
Dan Ehrenberg
2016/12/02 23:35:08
I think we would crash if out of memory, and the o
Yang
2016/12/05 19:19:30
Sounds right to me as well. No need to check for e
| |
1166 | 1195 |
1167 int sharp_s_count; | 1196 int sharp_s_count; |
1168 bool is_result_single_byte; | 1197 bool is_result_single_byte; |
1169 { | 1198 { |
1170 DisallowHeapAllocation no_gc; | 1199 DisallowHeapAllocation no_gc; |
1171 String::FlatContent flat = s->GetFlatContent(); | 1200 String::FlatContent flat = s->GetFlatContent(); |
1172 // If it was ok to slow down ASCII-only input slightly, ToUpperFastASCII | |
1173 // could be removed because ToUpperOneByte is pretty fast now (it | |
1174 // does not call ICU API any more.). | |
1175 if (flat.IsOneByte()) { | 1201 if (flat.IsOneByte()) { |
1176 Vector<const uint8_t> src = flat.ToOneByteVector(); | 1202 Vector<const uint8_t> src = flat.ToOneByteVector(); |
1177 if (ToUpperFastASCII(src, result)) return *result; | 1203 bool has_changed_character = false; |
1204 bool is_ascii = | |
1205 FastAsciiConvert<false>(reinterpret_cast<char*>(result->GetChars()), | |
1206 reinterpret_cast<const char*>(src.start()), | |
1207 length, &has_changed_character); | |
1208 // If not ASCII, we discard the result and use the table for Latin1. | |
1209 if (is_ascii) return has_changed_character ? *result : *s; | |
1178 is_result_single_byte = ToUpperOneByte(src, result, &sharp_s_count); | 1210 is_result_single_byte = ToUpperOneByte(src, result, &sharp_s_count); |
1179 } else { | 1211 } else { |
1180 DCHECK(flat.IsTwoByte()); | 1212 DCHECK(flat.IsTwoByte()); |
1181 Vector<const uint16_t> src = flat.ToUC16Vector(); | 1213 Vector<const uint16_t> src = flat.ToUC16Vector(); |
1182 if (ToUpperFastASCII(src, result)) return *result; | 1214 if (ToUpperFastASCII(src, result)) return *result; |
1183 is_result_single_byte = ToUpperOneByte(src, result, &sharp_s_count); | 1215 is_result_single_byte = ToUpperOneByte(src, result, &sharp_s_count); |
1184 } | 1216 } |
1185 } | 1217 } |
1186 | 1218 |
1187 // Go to the full Unicode path if there are characters whose uppercase | 1219 // Go to the full Unicode path if there are characters whose uppercase |
(...skipping 59 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... | |
1247 Handle<FixedArray> date_cache_version = | 1279 Handle<FixedArray> date_cache_version = |
1248 Handle<FixedArray>::cast(isolate->eternal_handles()->GetSingleton( | 1280 Handle<FixedArray>::cast(isolate->eternal_handles()->GetSingleton( |
1249 EternalHandles::DATE_CACHE_VERSION)); | 1281 EternalHandles::DATE_CACHE_VERSION)); |
1250 return date_cache_version->get(0); | 1282 return date_cache_version->get(0); |
1251 } | 1283 } |
1252 | 1284 |
1253 } // namespace internal | 1285 } // namespace internal |
1254 } // namespace v8 | 1286 } // namespace v8 |
1255 | 1287 |
1256 #endif // V8_I18N_SUPPORT | 1288 #endif // V8_I18N_SUPPORT |
OLD | NEW |