Migrate some case conversion functions from JS to CPP builtins

src/builtins/builtins-intl.cc

Index: src/builtins/builtins-intl.cc
diff --git a/src/builtins/builtins-intl.cc b/src/builtins/builtins-intl.cc
new file mode 100644
index 0000000000000000000000000000000000000000..5ef5841e14bbad3df8f961b02e0ec7a0ae356615
--- /dev/null
+++ b/src/builtins/builtins-intl.cc
@@ -0,0 +1,197 @@
+// Copyright 2017 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.
+
+#include "src/builtins/builtins-regexp.h"
+#include "src/builtins/builtins-utils.h"
+#include "src/builtins/builtins.h"
+#include "src/code-factory.h"
+#include "src/code-stub-assembler.h"
+#include "src/regexp/regexp-utils.h"
+#include "src/string-case.h"
+#include "src/unicode-inl.h"
+#include "src/unicode.h"
+
+namespace v8 {
+namespace internal {
+
+namespace {
+
+inline bool ToUpperOverflows(uc32 character) {
+  // y with umlauts and the micro sign are the only characters that stop
+  // fitting into one-byte when converting to uppercase.
+  static const uc32 yuml_code = 0xff;
+  static const uc32 micro_code = 0xb5;
+  return (character == yuml_code || character == micro_code);
+}
+
+template <class Converter>
+MUST_USE_RESULT static Object* ConvertCaseHelper(
+    Isolate* isolate, String* string, SeqString* result, int result_length,
+    unibrow::Mapping<Converter, 128>* mapping) {
+  DisallowHeapAllocation no_gc;
+  // We try this twice, once with the assumption that the result is no longer
+  // than the input and, if that assumption breaks, again with the exact
+  // length.  This may not be pretty, but it is nicer than what was here before
+  // and I hereby claim my vaffel-is.
+  //
+  // NOTE: This assumes that the upper/lower case of an ASCII
+  // character is also ASCII.  This is currently the case, but it
+  // might break in the future if we implement more context and locale
+  // dependent upper/lower conversions.
+  bool has_changed_character = false;
+
+  // Convert all characters to upper case, assuming that they will fit
+  // in the buffer
+  StringCharacterStream stream(string);
+  unibrow::uchar chars[Converter::kMaxWidth];
+  // We can assume that the string is not empty
+  uc32 current = stream.GetNext();
+  bool ignore_overflow = Converter::kIsToLower || result->IsSeqTwoByteString();
+  for (int i = 0; i < result_length;) {
+    bool has_next = stream.HasMore();
+    uc32 next = has_next ? stream.GetNext() : 0;
+    int char_length = mapping->get(current, next, chars);
+    if (char_length == 0) {
+      // The case conversion of this character is the character itself.
+      result->Set(i, current);
+      i++;
+    } else if (char_length == 1 &&
+               (ignore_overflow || !ToUpperOverflows(current))) {
+      // Common case: converting the letter resulted in one character.
+      DCHECK(static_cast<uc32>(chars[0]) != current);
+      result->Set(i, chars[0]);
+      has_changed_character = true;
+      i++;
+    } else if (result_length == string->length()) {
+      bool overflows = ToUpperOverflows(current);
+      // We've assumed that the result would be as long as the
+      // input but here is a character that converts to several
+      // characters.  No matter, we calculate the exact length
+      // of the result and try the whole thing again.
+      //
+      // Note that this leaves room for optimization.  We could just
+      // memcpy what we already have to the result string.  Also,
+      // the result string is the last object allocated we could
+      // "realloc" it and probably, in the vast majority of cases,
+      // extend the existing string to be able to hold the full
+      // result.
+      int next_length = 0;
+      if (has_next) {
+        next_length = mapping->get(next, 0, chars);
+        if (next_length == 0) next_length = 1;
+      }
+      int current_length = i + char_length + next_length;
+      while (stream.HasMore()) {
+        current = stream.GetNext();
+        overflows |= ToUpperOverflows(current);
+        // NOTE: we use 0 as the next character here because, while
+        // the next character may affect what a character converts to,
+        // it does not in any case affect the length of what it convert
+        // to.
+        int char_length = mapping->get(current, 0, chars);
+        if (char_length == 0) char_length = 1;
+        current_length += char_length;
+        if (current_length > String::kMaxLength) {
+          AllowHeapAllocation allocate_error_and_return;
+          THROW_NEW_ERROR_RETURN_FAILURE(isolate,
+                                         NewInvalidStringLengthError());
+        }
+      }
+      // Try again with the real length.  Return signed if we need
+      // to allocate a two-byte string for to uppercase.
+      return (overflows && !ignore_overflow) ? Smi::FromInt(-current_length)
+                                             : Smi::FromInt(current_length);
+    } else {
+      for (int j = 0; j < char_length; j++) {
+        result->Set(i, chars[j]);
+        i++;
+      }
+      has_changed_character = true;
+    }
+    current = next;
+  }
+  if (has_changed_character) {
+    return result;
+  } else {
+    // If we didn't actually change anything in doing the conversion
+    // we simple return the result and let the converted string
+    // become garbage; there is no reason to keep two identical strings
+    // alive.
+    return string;
+  }
+}
+
+template <class Converter>
+MUST_USE_RESULT static Object* ConvertCase(

[Dan Ehrenberg, 2017/03/14 16:19:56]

It looks like this case conversion code was copied from
src/builtins/builtins-string.cc . Instead, we should be using the case
conversion code in src/runtime/runtime-i18n.cc . This ConvertCase function has
the same name, but it does something different--calling out to ICU rather than
our own, custom case conversion ("unibrow").

+    Handle<String> s, Isolate* isolate,
+    unibrow::Mapping<Converter, 128>* mapping) {
+  s = String::Flatten(s);
+  int length = s->length();
+  // Assume that the string is not empty; we need this assumption later
+  if (length == 0) return *s;
+
+  // Simpler handling of ASCII strings.
+  //
+  // NOTE: This assumes that the upper/lower case of an ASCII
+  // character is also ASCII.  This is currently the case, but it
+  // might break in the future if we implement more context and locale
+  // dependent upper/lower conversions.
+  if (s->IsOneByteRepresentationUnderneath()) {
+    // Same length as input.
+    Handle<SeqOneByteString> result =
+        isolate->factory()->NewRawOneByteString(length).ToHandleChecked();
+    DisallowHeapAllocation no_gc;
+    String::FlatContent flat_content = s->GetFlatContent();
+    DCHECK(flat_content.IsFlat());
+    bool has_changed_character = false;
+    int index_to_first_unprocessed = FastAsciiConvert<Converter::kIsToLower>(
+        reinterpret_cast<char*>(result->GetChars()),
+        reinterpret_cast<const char*>(flat_content.ToOneByteVector().start()),
+        length, &has_changed_character);
+    // If not ASCII, we discard the result and take the 2 byte path.
+    if (index_to_first_unprocessed == length)
+      return has_changed_character ? *result : *s;
+  }
+
+  Handle<SeqString> result;  // Same length as input.
+  if (s->IsOneByteRepresentation()) {
+    result = isolate->factory()->NewRawOneByteString(length).ToHandleChecked();
+  } else {
+    result = isolate->factory()->NewRawTwoByteString(length).ToHandleChecked();
+  }
+
+  Object* answer = ConvertCaseHelper(isolate, *s, *result, length, mapping);
+  if (answer->IsException(isolate) || answer->IsString()) return answer;
+
+  DCHECK(answer->IsSmi());
+  length = Smi::cast(answer)->value();
+  if (s->IsOneByteRepresentation() && length > 0) {
+    ASSIGN_RETURN_FAILURE_ON_EXCEPTION(
+        isolate, result, isolate->factory()->NewRawOneByteString(length));
+  } else {
+    if (length < 0) length = -length;
+    ASSIGN_RETURN_FAILURE_ON_EXCEPTION(
+        isolate, result, isolate->factory()->NewRawTwoByteString(length));
+  }
+  return ConvertCaseHelper(isolate, *s, *result, length, mapping);
+}
+
+}  // namespace
+
+BUILTIN(StringPrototypeToLowerCaseI18N) {
+  HandleScope scope(isolate);
+  TO_THIS_STRING(string, "String.prototype.toLowerCase");
+  return ConvertCase(string, isolate,
+                     isolate->runtime_state()->to_lower_mapping());
+}
+
+BUILTIN(StringPrototypeToUpperCaseI18N) {
+  HandleScope scope(isolate);
+  TO_THIS_STRING(string, "String.prototype.toUpperCase");
+  return ConvertCase(string, isolate,
+                     isolate->runtime_state()->to_upper_mapping());
+}
+
+}  // namespace internal
+}  // namespace v8