Scanner for UTF-8 byte arrays

sdk/lib/_internal/compiler/implementation/scanner/token.dart

 // Copyright (c) 2011, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
 // BSD-style license that can be found in the LICENSE file.
 
 part of scanner;
 
 const int EOF_TOKEN = 0;
 
 const int KEYWORD_TOKEN = $k;
 const int IDENTIFIER_TOKEN = $a;
@@ skipping 61 matching lines @@
 const int TILDE_SLASH_TOKEN = TILDE_SLASH_EQ_TOKEN + 1;
 const int PERCENT_EQ_TOKEN = TILDE_SLASH_TOKEN + 1;
 const int GT_GT_TOKEN = PERCENT_EQ_TOKEN + 1;
 const int CARET_EQ_TOKEN = GT_GT_TOKEN + 1;
 const int COMMENT_TOKEN = CARET_EQ_TOKEN + 1;
 const int STRING_INTERPOLATION_IDENTIFIER_TOKEN = COMMENT_TOKEN + 1;
 
 /**
  * A token that doubles as a linked list.
  */
-class Token implements Spannable {
-  /**
-   * The precedence info for this token. [info] determines the kind and the
-   * precedence level of this token.
-   */
-  final PrecedenceInfo info;
-
+abstract class Token implements Spannable {
   /**
    * The character offset of the start of this token within the source text.
    */
   final int charOffset;
 
+  Token(this.charOffset);
+
   /**
    * The next token in the token stream.
    */
   Token next;
 
-  Token(this.info, this.charOffset);
-
-  get value => info.value;
+  /**
+   * The precedence info for this token. [info] determines the kind and the
+   * precedence level of this token.
+   *
+   * Defined as getter to save a field in the [KeywordToken] subclass.
+   */
+  PrecedenceInfo get info;
 
   /**
-   * Returns the string value for keywords and symbols. For instance 'class' for
-   * the [CLASS] keyword token and '*' for a [Token] based on [STAR_INFO]. For
-   * other tokens, such identifiers, strings, numbers, etc, [stringValue]
-   * returns [:null:].
+   * The string represented by this token, a substring of the source code.
    *
-   * [stringValue] should only be used for testing keywords and symbols.
+   * For [StringToken]s the value includes the quotes, explicit escapes, etc.
+   *
    */
-  String get stringValue => info.value.stringValue;
+  String get value;
+
+  /**
+   * For symbol and keyword tokens, returns the string value reprenseted by this
+   * token. For [StringToken]s this method returns [:null:].
+   *
+   * For [SymbolToken]s and [KeywordToken]s, the string value is a compile-time
+   * constant originating in the [PrecedenceInfo] or in the [Keyword] instance.
+   * This allows testing for keywords and symbols using [:identical:], e.g.,
+   * [:identical('class', token.value):].
+   *
+   * Note that returning [:null:] for string tokens is important to identify
+   * symbols and keywords, we cannot use [value] instead. The string literal
+   *   "$a($b"
+   * produces ..., SymbolToken($), StringToken(a), StringToken((), ...
+   *
+   * After parsing the identifier 'a', the parser tests for a function
+   * declaration using [:identical(next.stringValue, '('):], which (rihgtfully)
+   * returns false because stringValue returns [:null:].
+   */
+  String get stringValue;
 
   /**
    * The kind enum of this token as determined by its [info].
    */
   int get kind => info.kind;
 
   /**
    * The precedence level for this token.
    */
   int get precedence => info.precedence;
 
-  bool isIdentifier() => identical(kind, IDENTIFIER_TOKEN);
+  /**
+   * True if this token is an identifier. Some keywords allowed as identifiers,
+   * see implementaiton in [KeywordToken].
+   */
+  bool isIdentifier();
 
   /**
    * Returns a textual representation of this token to be used for debugging
    * purposes. The resulting string might contain information about the
    * structure of the token, for example 'StringToken(foo)' for the identifier
-   * token 'foo'. Use [slowToString] for the text actually parsed by the token.
+   * token 'foo'.
+   *
+   * Use [value] for the text actually parsed by the token.
    */
-  String toString() => info.value.toString();
-
-  /**
-   * The text parsed by this token.
-   */
-  String slowToString() => toString();
+  String toString();
 
   /**
    * The number of characters parsed by this token.
    */
-  int get slowCharCount {
+  int get charCount {
     if (info == BAD_INPUT_INFO) {
       // This is a token that wraps around an error message. Return 1
       // instead of the size of the length of the error message.
       return 1;
     } else {
-      return slowToString().length;
+      return value.length;
     }
   }
 
   int get hashCode => computeHashCode(charOffset, info, value);
 }
 
 /**
+ * A symbol token represents the symbol in its precendence info.
+ * Also used for end of file with EOF_INFO.
+ */
+class SymbolToken extends Token {
+
+  final PrecedenceInfo info;
+
+  SymbolToken(this.info, int charOffset) : super(charOffset);
+
+  String get value => info.value;
+
+  String get stringValue => info.value;
+
+  bool isIdentifier() => false;
+
+  String toString() => "SymbolToken($value)";
+}
+
+/**
+ * A [BeginGroupToken] reprsents a symbol that may be the beginning of
+ * a pair of brackets, i.e., ( { [ < or ${
+ * The [endGroup] token points to the matching closing bracked in case
+ * it can be identified during scanning.
+ */
+class BeginGroupToken extends SymbolToken {
+  Token endGroup;
+
+  BeginGroupToken(PrecedenceInfo info, int charOffset)
+    : super(info, charOffset);
+}
+
+/**
  * A keyword token.
  */
 class KeywordToken extends Token {
-  final Keyword value;
-  String get stringValue => value.syntax;
-
-  KeywordToken(Keyword value, int charOffset)
-    : this.value = value, super(value.info, charOffset);
-
-  bool isIdentifier() => value.isPseudo || value.isBuiltIn;
-
-  String toString() => value.syntax;
-}
-
-/**
- * A String-valued token.
+  final Keyword keyword;
+
+  KeywordToken(this.keyword, int charOffset) : super(charOffset);
+
+  PrecedenceInfo get info => keyword.info;
+
+  String get value => keyword.syntax;
+
+  String get stringValue => keyword.syntax;
+
+  bool isIdentifier() => keyword.isPseudo || keyword.isBuiltIn;
+
+  String toString() => "KeywordToken($value)";
+}
+
+/**
+ * A String-valued token. Represents identifiers, string literals,
+ * number literals, comments and error tokens, using the corresponding
+ * precedence info.
  */
 class StringToken extends Token {
-  final SourceString value;
-
-  StringToken(PrecedenceInfo info, String value, int charOffset)
-    : this.fromSource(info, new SourceString(value), charOffset);
-
-  StringToken.fromSource(PrecedenceInfo info, this.value, int charOffset)
-    : super(info, charOffset);
-
-  String toString() => "StringToken(${value.slowToString()})";
-
-  String slowToString() => value.slowToString();
-}
-
-abstract class SourceString extends IterableBase<int> {
-  const factory SourceString(String string) = StringWrapper;
-
-  static final Map<String, StringWrapper> canonicalizedValues =
-      new Map<String, StringWrapper>();
-
-  factory SourceString.fromSubstring(String string, int begin, int end) {
-    var substring = string.substring(begin, end);
-    return canonicalizedValues.putIfAbsent(
-        substring, () => new StringWrapper(substring));
-  }
-
-  void printOn(StringBuffer sb);
-
-  /** Gives a [SourceString] that is not including the [initial] first and
-   * [terminal] last characters. This is only intended to be used to remove
-   * quotes from string literals (including an initial '@' for raw strings).
-   */
-  SourceString copyWithoutQuotes(int initial, int terminal);
-
-  String get stringValue;
-
-  String slowToString();
-
-  bool get isEmpty;
-
-  bool isPrivate();
-}
-
-class StringWrapper extends IterableBase<int> implements SourceString {
-  final String stringValue;
-
-  const StringWrapper(this.stringValue);
-
-  int get hashCode => stringValue.hashCode;
-
-  bool operator ==(other) {
-    return other is SourceString && toString() == other.slowToString();
-  }
-
-  Iterator<int> get iterator => new StringCodeIterator(stringValue);
-
-  void printOn(StringBuffer sb) {
-    sb.write(stringValue);
-  }
-
-  String toString() => stringValue;
-
-  String slowToString() => stringValue;
-
-  SourceString copyWithoutQuotes(int initial, int terminal) {
-    assert(0 <= initial);
-    assert(0 <= terminal);
-    assert(initial + terminal <= stringValue.length);
-    return new StringWrapper(
-        stringValue.substring(initial, stringValue.length - terminal));
-  }
-
-  bool get isEmpty => stringValue.isEmpty;
-
-  bool isPrivate() => !isEmpty && stringValue.codeUnitAt(0) == $_;
-}
-
-class StringCodeIterator implements Iterator<int> {
-  final String string;
-  int index;
-  final int end;
-  int _current;
-
-  StringCodeIterator(String string) :
-    this.string = string, index = 0, end = string.length;
-
-  StringCodeIterator.substring(this.string, this.index, this.end) {
-    assert(0 <= index);
-    assert(index <= end);
-    assert(end <= string.length);
-  }
-
-  int get current => _current;
-
-  bool moveNext() {
-    _current = null;
-    if (index >= end) return false;
-    _current = string.codeUnitAt(index++);
-    return true;
-  }
-}
-
-class BeginGroupToken extends StringToken {
-  Token endGroup;
-  BeginGroupToken(PrecedenceInfo info, String value, int charOffset)
-    : super(info, value, charOffset);
+  /**
+   * The length thershold above which substring tokens are computed lazily.

[kasperl, 2013/10/17 08:50:39]

thershold -> threshold

[lukas, 2013/10/17 17:49:34]

Done.

+   *
+   * For string tokens that are substrings of the program source, the actual
+   * substring extraction is performed lazily. This is beneficial because
+   * not all scanned code is actually used. For unused parts, the substrings
+   * are never computed and allocated.
+   */
+  static final int lazyThreshold = 4;

[kasperl, 2013/10/17 08:50:39]

static const int LAZY_THRESHOLD

[lukas, 2013/10/17 17:49:34]

Done.

+
+  var valueOrSublist;

[kasperl, 2013/10/17 08:50:39]

Sublist -> LazyList?

[lukas, 2013/10/17 17:49:34]

Done.

+
+  final PrecedenceInfo info;
+
+  /**
+   * Creates a non-lazy string token. If [canonicalize] is true, the string
+   * is canonicalized before the token is created.
+   */
+  StringToken.fromString(this.info, String value, int charOffset,
+                         [bool canonicalize = false])
+      : valueOrSublist = canonicalizedString(value, canonicalize),
+        super(charOffset);
+
+  /**
+   * Creates a lazy string token. If [canonicalize] is true, the string
+   * is canonicalized before the token is created.
+   */
+  StringToken.fromSubstring(this.info, String data, int start, int end,
+                            int charOffset, [bool canonicalize = false])
+      : super(charOffset) {
+    int length = end - start;
+    if (length <= lazyThreshold) {
+      valueOrSublist = canonicalizedString(data.substring(start, end),
+                                            canonicalize);
+    } else {
+      valueOrSublist = new LazySubstring(data, start, length, canonicalize);
+    }
+  }
+
+  /**
+   * Creates a lazy string token. If [asciiOnly] is false, the byte array
+   * is passed through a UTF-8 decoder.
+   */
+  StringToken.fromUtf8Bytes(this.info, List<int> data, int start, int end,
+                            bool asciiOnly, int charOffset)
+      : super(charOffset) {
+    int length = end - start;
+    if (length <= lazyThreshold) {
+      valueOrSublist = decodeUtf8(data, start, end, asciiOnly);
+    } else {
+      valueOrSublist = new LazySubstring(data, start, length, asciiOnly);
+    }
+  }
+
+  String get value {
+    if (valueOrSublist is String) {
+      return valueOrSublist;
+    } else {
+      if (valueOrSublist is LazySubstring) {

[kasperl, 2013/10/17 08:50:39]

Should this be an assert instead?

[lukas, 2013/10/17 17:49:34]

Done.

+        var data = valueOrSublist.data;
+        int start = valueOrSublist.start;
+        int end = start + valueOrSublist.length;
+        if (data is String) {
+          valueOrSublist = canonicalizedString(data.substring(start, end),
+                                               valueOrSublist.boolValue);
+        } else {
+          valueOrSublist = decodeUtf8(data, start, end,
+                                      valueOrSublist.boolValue);
+        }
+      }
+      return valueOrSublist;
+    }
+  }
+
+  String get stringValue => null;
+
+  bool isIdentifier() => identical(kind, IDENTIFIER_TOKEN);
+
+  String toString() => "StringToken($value)";
+
+  // @lry replace by hash set after merging from svn trunk

[kasperl, 2013/10/17 08:50:39]

I guess this is done?

[lukas, 2013/10/17 17:49:34]

Done.

+  static final HashSet<String> canonicalizedSubstrings =
+      new HashSet();
+
+  static String canonicalizedString(String s, bool canonicalize) {
+    if (canonicalize) {

[kasperl, 2013/10/17 08:50:39]

I'd consider doing something like:

  if (!canonicalize) return s;
  ...

to get rid of some of the nesting.

[lukas, 2013/10/17 17:49:34]

Done.

+      var result = canonicalizedSubstrings.lookup(s);
+      if (result == null) {

[kasperl, 2013/10/17 08:50:39]

I'd consider:

  if (result != null) return result;

to get rid of even more nesting.

[lukas, 2013/10/17 17:49:34]

Done.

+        canonicalizedSubstrings.add(s);
+        return s;
+      } else {
+        return result;
+      }
+    } else {
+      return s;
+    }
+  }
+
+  static String decodeUtf8(List<int> data, int start, int end, bool asciiOnly) {
+    var s;
+    if (asciiOnly) {
+      s = new String.fromCharCodes(data.getRange(start, end));

[kasperl, 2013/10/17 08:50:39]

Maybe add a comment that getRange doesn't copy the data, but it gives an
iterable "view" of the range.

[lukas, 2013/10/17 17:49:34]

Done.

+    } else {
+      // TODO(lry), this is measurably slow. Also sublist is allocated eagerly.

[kasperl, 2013/10/17 08:50:39]

TODO(lry), -> TODO(lry):. Instead of allocated you might use a word like copy or
something.

[lukas, 2013/10/17 17:49:34]

Done.

+      var bytes = data.sublist(start, end);
+      s = UTF8.decode(bytes);
+    }
+    return canonicalizedString(s, true);
+  }
+}
+
+/**
+ * This class represents the necessary information to compute a substring
+ * lazily. The substring can either originate in a string or in a [:List<int>:]
+ * of UTF-8 bytes.
+ */
+abstract class LazySubstring {
+  /** The original data, either a string or a List<int> */
+  get data;
+
+  int get start;
+  int get length;
+
+  /**
+   * If this substring is based on a String, the boolean indicates wheter the
+   * resulting substring should be canonicalized.
+   *
+   * For substrings based on a byte array, the boolean value is true if the
+   * array only holds ASCII characters. The resulting substring will be
+   * canonicalized after decoding.
+   */
+  bool get boolValue;
+
+  LazySubstring.internal();
+
+  factory LazySubstring(data, int start, int length, bool b) {
+    // See comment on [CompactLazySubstring].
+    if (start < 0x100000 && length < 0x200) {
+      int fields = (start << 9);
+      fields = fields | length;
+      fields = fields << 1;
+      if (b) fields |= 1;
+      return new CompactLazySubstring(data, fields);
+    } else {
+      return new FullLazySubstring(data, start, length, b);
+    }
+  }
+}
+
+/**
+ * This class encodes [start], [length] and [boolValue] in a single
+ * 30 bit integer. It uses 20 bits for [start], which covers source files
+ * of 1M. [length] has 9 bits, which covers 512 characters.
+ *
+ * The file html_dart2js.dart is currently around 1M.
+ */
+class CompactLazySubstring extends LazySubstring {
+  final data;
+  final int fields;
+
+  CompactLazySubstring(this.data, this.fields) : super.internal();
+
+  int get start => fields >> 10;
+  int get length => (fields >> 1) & 0x1ff;
+  bool get boolValue => (fields & 1) == 1;
+}
+
+class FullLazySubstring extends LazySubstring {
+  final data;
+  final int start;
+  final int length;
+  final bool boolValue;
+  FullLazySubstring(this.data, this.start, this.length, this.boolValue)
+      : super.internal();
 }
 
 bool isUserDefinableOperator(String value) {
   return
       isBinaryOperator(value) ||
       isMinusOperator(value) ||
       isTernaryOperator(value) ||
       isUnaryOperator(value);
 }
 
@@ skipping 250 matching lines @@
 
 const PrecedenceInfo STRING_INTERPOLATION_IDENTIFIER_INFO =
   const PrecedenceInfo('\$', 0,
                        STRING_INTERPOLATION_IDENTIFIER_TOKEN);
 
 const PrecedenceInfo HEXADECIMAL_INFO =
   const PrecedenceInfo('hexadecimal', 0, HEXADECIMAL_TOKEN);
 
 const PrecedenceInfo COMMENT_INFO =
   const PrecedenceInfo('comment', 0, COMMENT_TOKEN);