Index: pkg/analyzer-experimental/lib/src/generated/scanner.dart |
diff --git a/pkg/analyzer-experimental/lib/src/generated/scanner.dart b/pkg/analyzer-experimental/lib/src/generated/scanner.dart |
deleted file mode 100644 |
index 931408f8970053e2e6b0f4f12ca512cd7c97408f..0000000000000000000000000000000000000000 |
--- a/pkg/analyzer-experimental/lib/src/generated/scanner.dart |
+++ /dev/null |
@@ -1,1887 +0,0 @@ |
-// This code was auto-generated, is not intended to be edited, and is subject to |
-// significant change. Please see the README file for more information. |
- |
-library engine.scanner; |
- |
-import 'dart:collection'; |
-import 'java_core.dart'; |
-import 'source.dart'; |
-import 'error.dart'; |
-import 'instrumentation.dart'; |
- |
-/** |
- * Instances of the abstract class {@code KeywordState} represent a state in a state machine used to |
- * scan keywords. |
- * @coverage dart.engine.parser |
- */ |
-class KeywordState { |
- /** |
- * An empty transition table used by leaf states. |
- */ |
- static List<KeywordState> _EMPTY_TABLE = new List<KeywordState>(26); |
- /** |
- * The initial state in the state machine. |
- */ |
- static KeywordState KEYWORD_STATE = createKeywordStateTable(); |
- /** |
- * Create the next state in the state machine where we have already recognized the subset of |
- * strings in the given array of strings starting at the given offset and having the given length. |
- * All of these strings have a common prefix and the next character is at the given start index. |
- * @param start the index of the character in the strings used to transition to a new state |
- * @param strings an array containing all of the strings that will be recognized by the state |
- * machine |
- * @param offset the offset of the first string in the array that has the prefix that is assumed |
- * to have been recognized by the time we reach the state being built |
- * @param length the number of strings in the array that pass through the state being built |
- * @return the state that was created |
- */ |
- static KeywordState computeKeywordStateTable(int start, List<String> strings, int offset, int length12) { |
- List<KeywordState> result = new List<KeywordState>(26); |
- assert(length12 != 0); |
- int chunk = 0x0; |
- int chunkStart = -1; |
- bool isLeaf = false; |
- for (int i = offset; i < offset + length12; i++) { |
- if (strings[i].length == start) { |
- isLeaf = true; |
- } |
- if (strings[i].length > start) { |
- int c = strings[i].codeUnitAt(start); |
- if (chunk != c) { |
- if (chunkStart != -1) { |
- result[chunk - 0x61] = computeKeywordStateTable(start + 1, strings, chunkStart, i - chunkStart); |
- } |
- chunkStart = i; |
- chunk = c; |
- } |
- } |
- } |
- if (chunkStart != -1) { |
- assert(result[chunk - 0x61] == null); |
- result[chunk - 0x61] = computeKeywordStateTable(start + 1, strings, chunkStart, offset + length12 - chunkStart); |
- } else { |
- assert(length12 == 1); |
- return new KeywordState(_EMPTY_TABLE, strings[offset]); |
- } |
- if (isLeaf) { |
- return new KeywordState(result, strings[offset]); |
- } else { |
- return new KeywordState(result, null); |
- } |
- } |
- /** |
- * Create the initial state in the state machine. |
- * @return the state that was created |
- */ |
- static KeywordState createKeywordStateTable() { |
- List<Keyword> values2 = Keyword.values; |
- List<String> strings = new List<String>(values2.length); |
- for (int i = 0; i < values2.length; i++) { |
- strings[i] = values2[i].syntax; |
- } |
- strings.sort(); |
- return computeKeywordStateTable(0, strings, 0, strings.length); |
- } |
- /** |
- * A table mapping characters to the states to which those characters will transition. (The index |
- * into the array is the offset from the character {@code 'a'} to the transitioning character.) |
- */ |
- List<KeywordState> _table; |
- /** |
- * The keyword that is recognized by this state, or {@code null} if this state is not a terminal |
- * state. |
- */ |
- Keyword _keyword2; |
- /** |
- * Initialize a newly created state to have the given transitions and to recognize the keyword |
- * with the given syntax. |
- * @param table a table mapping characters to the states to which those characters will transition |
- * @param syntax the syntax of the keyword that is recognized by the state |
- */ |
- KeywordState(List<KeywordState> table, String syntax) { |
- this._table = table; |
- this._keyword2 = (syntax == null) ? null : Keyword.keywords[syntax]; |
- } |
- /** |
- * Return the keyword that was recognized by this state, or {@code null} if this state does not |
- * recognized a keyword. |
- * @return the keyword that was matched by reaching this state |
- */ |
- Keyword keyword() => _keyword2; |
- /** |
- * Return the state that follows this state on a transition of the given character, or{@code null} if there is no valid state reachable from this state with such a transition. |
- * @param c the character used to transition from this state to another state |
- * @return the state that follows this state on a transition of the given character |
- */ |
- KeywordState next(int c) => _table[c - 0x61]; |
-} |
-/** |
- * The enumeration {@code ScannerErrorCode} defines the error codes used for errors detected by the |
- * scanner. |
- * @coverage dart.engine.parser |
- */ |
-class ScannerErrorCode implements ErrorCode { |
- static final ScannerErrorCode ILLEGAL_CHARACTER = new ScannerErrorCode('ILLEGAL_CHARACTER', 0, "Illegal character %x"); |
- static final ScannerErrorCode MISSING_DIGIT = new ScannerErrorCode('MISSING_DIGIT', 1, "Decimal digit expected"); |
- static final ScannerErrorCode MISSING_HEX_DIGIT = new ScannerErrorCode('MISSING_HEX_DIGIT', 2, "Hexidecimal digit expected"); |
- static final ScannerErrorCode MISSING_QUOTE = new ScannerErrorCode('MISSING_QUOTE', 3, "Expected quote (' or \")"); |
- static final ScannerErrorCode UNTERMINATED_MULTI_LINE_COMMENT = new ScannerErrorCode('UNTERMINATED_MULTI_LINE_COMMENT', 4, "Unterminated multi-line comment"); |
- static final ScannerErrorCode UNTERMINATED_STRING_LITERAL = new ScannerErrorCode('UNTERMINATED_STRING_LITERAL', 5, "Unterminated string literal"); |
- static final List<ScannerErrorCode> values = [ILLEGAL_CHARACTER, MISSING_DIGIT, MISSING_HEX_DIGIT, MISSING_QUOTE, UNTERMINATED_MULTI_LINE_COMMENT, UNTERMINATED_STRING_LITERAL]; |
- final String __name; |
- final int __ordinal; |
- int get ordinal => __ordinal; |
- /** |
- * The message template used to create the message to be displayed for this error. |
- */ |
- String _message; |
- /** |
- * Initialize a newly created error code to have the given message. |
- * @param message the message template used to create the message to be displayed for this error |
- */ |
- ScannerErrorCode(this.__name, this.__ordinal, String message) { |
- this._message = message; |
- } |
- ErrorSeverity get errorSeverity => ErrorSeverity.ERROR; |
- String get message => _message; |
- ErrorType get type => ErrorType.SYNTACTIC_ERROR; |
- bool needsRecompilation() => true; |
- String toString() => __name; |
-} |
-/** |
- * Instances of the class {@code TokenWithComment} represent a string token that is preceded by |
- * comments. |
- * @coverage dart.engine.parser |
- */ |
-class StringTokenWithComment extends StringToken { |
- /** |
- * The first comment in the list of comments that precede this token. |
- */ |
- Token _precedingComment; |
- /** |
- * Initialize a newly created token to have the given type and offset and to be preceded by the |
- * comments reachable from the given comment. |
- * @param type the type of the token |
- * @param offset the offset from the beginning of the file to the first character in the token |
- * @param precedingComment the first comment in the list of comments that precede this token |
- */ |
- StringTokenWithComment(TokenType type, String value, int offset, Token precedingComment) : super(type, value, offset) { |
- this._precedingComment = precedingComment; |
- } |
- Token get precedingComments => _precedingComment; |
-} |
-/** |
- * The enumeration {@code Keyword} defines the keywords in the Dart programming language. |
- * @coverage dart.engine.parser |
- */ |
-class Keyword { |
- static final Keyword ASSERT = new Keyword.con1('ASSERT', 0, "assert"); |
- static final Keyword BREAK = new Keyword.con1('BREAK', 1, "break"); |
- static final Keyword CASE = new Keyword.con1('CASE', 2, "case"); |
- static final Keyword CATCH = new Keyword.con1('CATCH', 3, "catch"); |
- static final Keyword CLASS = new Keyword.con1('CLASS', 4, "class"); |
- static final Keyword CONST = new Keyword.con1('CONST', 5, "const"); |
- static final Keyword CONTINUE = new Keyword.con1('CONTINUE', 6, "continue"); |
- static final Keyword DEFAULT = new Keyword.con1('DEFAULT', 7, "default"); |
- static final Keyword DO = new Keyword.con1('DO', 8, "do"); |
- static final Keyword ELSE = new Keyword.con1('ELSE', 9, "else"); |
- static final Keyword EXTENDS = new Keyword.con1('EXTENDS', 10, "extends"); |
- static final Keyword FALSE = new Keyword.con1('FALSE', 11, "false"); |
- static final Keyword FINAL = new Keyword.con1('FINAL', 12, "final"); |
- static final Keyword FINALLY = new Keyword.con1('FINALLY', 13, "finally"); |
- static final Keyword FOR = new Keyword.con1('FOR', 14, "for"); |
- static final Keyword IF = new Keyword.con1('IF', 15, "if"); |
- static final Keyword IN = new Keyword.con1('IN', 16, "in"); |
- static final Keyword IS = new Keyword.con1('IS', 17, "is"); |
- static final Keyword NEW = new Keyword.con1('NEW', 18, "new"); |
- static final Keyword NULL = new Keyword.con1('NULL', 19, "null"); |
- static final Keyword RETURN = new Keyword.con1('RETURN', 20, "return"); |
- static final Keyword SUPER = new Keyword.con1('SUPER', 21, "super"); |
- static final Keyword SWITCH = new Keyword.con1('SWITCH', 22, "switch"); |
- static final Keyword THIS = new Keyword.con1('THIS', 23, "this"); |
- static final Keyword THROW = new Keyword.con1('THROW', 24, "throw"); |
- static final Keyword TRUE = new Keyword.con1('TRUE', 25, "true"); |
- static final Keyword TRY = new Keyword.con1('TRY', 26, "try"); |
- static final Keyword VAR = new Keyword.con1('VAR', 27, "var"); |
- static final Keyword VOID = new Keyword.con1('VOID', 28, "void"); |
- static final Keyword WHILE = new Keyword.con1('WHILE', 29, "while"); |
- static final Keyword WITH = new Keyword.con1('WITH', 30, "with"); |
- static final Keyword ABSTRACT = new Keyword.con2('ABSTRACT', 31, "abstract", true); |
- static final Keyword AS = new Keyword.con2('AS', 32, "as", true); |
- static final Keyword DYNAMIC = new Keyword.con2('DYNAMIC', 33, "dynamic", true); |
- static final Keyword EXPORT = new Keyword.con2('EXPORT', 34, "export", true); |
- static final Keyword EXTERNAL = new Keyword.con2('EXTERNAL', 35, "external", true); |
- static final Keyword FACTORY = new Keyword.con2('FACTORY', 36, "factory", true); |
- static final Keyword GET = new Keyword.con2('GET', 37, "get", true); |
- static final Keyword IMPLEMENTS = new Keyword.con2('IMPLEMENTS', 38, "implements", true); |
- static final Keyword IMPORT = new Keyword.con2('IMPORT', 39, "import", true); |
- static final Keyword LIBRARY = new Keyword.con2('LIBRARY', 40, "library", true); |
- static final Keyword OPERATOR = new Keyword.con2('OPERATOR', 41, "operator", true); |
- static final Keyword PART = new Keyword.con2('PART', 42, "part", true); |
- static final Keyword SET = new Keyword.con2('SET', 43, "set", true); |
- static final Keyword STATIC = new Keyword.con2('STATIC', 44, "static", true); |
- static final Keyword TYPEDEF = new Keyword.con2('TYPEDEF', 45, "typedef", true); |
- static final List<Keyword> values = [ASSERT, BREAK, CASE, CATCH, CLASS, CONST, CONTINUE, DEFAULT, DO, ELSE, EXTENDS, FALSE, FINAL, FINALLY, FOR, IF, IN, IS, NEW, NULL, RETURN, SUPER, SWITCH, THIS, THROW, TRUE, TRY, VAR, VOID, WHILE, WITH, ABSTRACT, AS, DYNAMIC, EXPORT, EXTERNAL, FACTORY, GET, IMPLEMENTS, IMPORT, LIBRARY, OPERATOR, PART, SET, STATIC, TYPEDEF]; |
- String __name; |
- int __ordinal = 0; |
- int get ordinal => __ordinal; |
- /** |
- * The lexeme for the keyword. |
- */ |
- String _syntax; |
- /** |
- * A flag indicating whether the keyword is a pseudo-keyword. Pseudo keywords can be used as |
- * identifiers. |
- */ |
- bool _isPseudoKeyword2 = false; |
- /** |
- * A table mapping the lexemes of keywords to the corresponding keyword. |
- */ |
- static Map<String, Keyword> keywords = createKeywordMap(); |
- /** |
- * Create a table mapping the lexemes of keywords to the corresponding keyword. |
- * @return the table that was created |
- */ |
- static Map<String, Keyword> createKeywordMap() { |
- LinkedHashMap<String, Keyword> result = new LinkedHashMap<String, Keyword>(); |
- for (Keyword keyword in values) { |
- result[keyword._syntax] = keyword; |
- } |
- return result; |
- } |
- /** |
- * Initialize a newly created keyword to have the given syntax. The keyword is not a |
- * pseudo-keyword. |
- * @param syntax the lexeme for the keyword |
- */ |
- Keyword.con1(String ___name, int ___ordinal, String syntax) { |
- _jtd_constructor_265_impl(___name, ___ordinal, syntax); |
- } |
- _jtd_constructor_265_impl(String ___name, int ___ordinal, String syntax) { |
- _jtd_constructor_266_impl(___name, ___ordinal, syntax, false); |
- } |
- /** |
- * Initialize a newly created keyword to have the given syntax. The keyword is a pseudo-keyword if |
- * the given flag is {@code true}. |
- * @param syntax the lexeme for the keyword |
- * @param isPseudoKeyword {@code true} if this keyword is a pseudo-keyword |
- */ |
- Keyword.con2(String ___name, int ___ordinal, String syntax2, bool isPseudoKeyword) { |
- _jtd_constructor_266_impl(___name, ___ordinal, syntax2, isPseudoKeyword); |
- } |
- _jtd_constructor_266_impl(String ___name, int ___ordinal, String syntax2, bool isPseudoKeyword) { |
- __name = ___name; |
- __ordinal = ___ordinal; |
- this._syntax = syntax2; |
- this._isPseudoKeyword2 = isPseudoKeyword; |
- } |
- /** |
- * Return the lexeme for the keyword. |
- * @return the lexeme for the keyword |
- */ |
- String get syntax => _syntax; |
- /** |
- * Return {@code true} if this keyword is a pseudo-keyword. Pseudo keywords can be used as |
- * identifiers. |
- * @return {@code true} if this keyword is a pseudo-keyword |
- */ |
- bool isPseudoKeyword() => _isPseudoKeyword2; |
- String toString() => __name; |
-} |
-/** |
- * The abstract class {@code AbstractScanner} implements a scanner for Dart code. Subclasses are |
- * required to implement the interface used to access the characters being scanned. |
- * <p> |
- * The lexical structure of Dart is ambiguous without knowledge of the context in which a token is |
- * being scanned. For example, without context we cannot determine whether source of the form "<<" |
- * should be scanned as a single left-shift operator or as two left angle brackets. This scanner |
- * does not have any context, so it always resolves such conflicts by scanning the longest possible |
- * token. |
- * @coverage dart.engine.parser |
- */ |
-abstract class AbstractScanner { |
- /** |
- * The source being scanned. |
- */ |
- Source _source; |
- /** |
- * The error listener that will be informed of any errors that are found during the scan. |
- */ |
- AnalysisErrorListener _errorListener; |
- /** |
- * The token pointing to the head of the linked list of tokens. |
- */ |
- Token _tokens; |
- /** |
- * The last token that was scanned. |
- */ |
- Token _tail; |
- /** |
- * The first token in the list of comment tokens found since the last non-comment token. |
- */ |
- Token _firstComment; |
- /** |
- * The last token in the list of comment tokens found since the last non-comment token. |
- */ |
- Token _lastComment; |
- /** |
- * The index of the first character of the current token. |
- */ |
- int _tokenStart = 0; |
- /** |
- * A list containing the offsets of the first character of each line in the source code. |
- */ |
- List<int> _lineStarts = new List<int>(); |
- /** |
- * A list, treated something like a stack, of tokens representing the beginning of a matched pair. |
- * It is used to pair the end tokens with the begin tokens. |
- */ |
- List<BeginToken> _groupingStack = new List<BeginToken>(); |
- /** |
- * A flag indicating whether any unmatched groups were found during the parse. |
- */ |
- bool _hasUnmatchedGroups2 = false; |
- /** |
- * A non-breaking space, which is allowed by this scanner as a white-space character. |
- */ |
- static int _$NBSP = 160; |
- /** |
- * Initialize a newly created scanner. |
- * @param source the source being scanned |
- * @param errorListener the error listener that will be informed of any errors that are found |
- */ |
- AbstractScanner(Source source, AnalysisErrorListener errorListener) { |
- this._source = source; |
- this._errorListener = errorListener; |
- _tokens = new Token(TokenType.EOF, -1); |
- _tokens.setNext(_tokens); |
- _tail = _tokens; |
- _tokenStart = -1; |
- _lineStarts.add(0); |
- } |
- /** |
- * Return an array containing the offsets of the first character of each line in the source code. |
- * @return an array containing the offsets of the first character of each line in the source code |
- */ |
- List<int> get lineStarts => _lineStarts; |
- /** |
- * Return the current offset relative to the beginning of the file. Return the initial offset if |
- * the scanner has not yet scanned the source code, and one (1) past the end of the source code if |
- * the source code has been scanned. |
- * @return the current offset of the scanner in the source |
- */ |
- int get offset; |
- /** |
- * Return {@code true} if any unmatched groups were found during the parse. |
- * @return {@code true} if any unmatched groups were found during the parse |
- */ |
- bool hasUnmatchedGroups() => _hasUnmatchedGroups2; |
- /** |
- * Scan the source code to produce a list of tokens representing the source. |
- * @return the first token in the list of tokens that were produced |
- */ |
- Token tokenize() { |
- int next = advance(); |
- while (next != -1) { |
- next = bigSwitch(next); |
- } |
- appendEofToken(); |
- return firstToken(); |
- } |
- /** |
- * Advance the current position and return the character at the new current position. |
- * @return the character at the new current position |
- */ |
- int advance(); |
- /** |
- * Return the substring of the source code between the start offset and the modified current |
- * position. The current position is modified by adding the end delta. |
- * @param start the offset to the beginning of the string, relative to the start of the file |
- * @param endDelta the number of character after the current location to be included in the |
- * string, or the number of characters before the current location to be excluded if the |
- * offset is negative |
- * @return the specified substring of the source code |
- */ |
- String getString(int start, int endDelta); |
- /** |
- * Return the character at the current position without changing the current position. |
- * @return the character at the current position |
- */ |
- int peek(); |
- /** |
- * Record the fact that we are at the beginning of a new line in the source. |
- */ |
- void recordStartOfLine() { |
- _lineStarts.add(offset); |
- } |
- void appendBeginToken(TokenType type) { |
- BeginToken token; |
- if (_firstComment == null) { |
- token = new BeginToken(type, _tokenStart); |
- } else { |
- token = new BeginTokenWithComment(type, _tokenStart, _firstComment); |
- _firstComment = null; |
- _lastComment = null; |
- } |
- _tail = _tail.setNext(token); |
- _groupingStack.add(token); |
- } |
- void appendCommentToken(TokenType type, String value) { |
- if (_firstComment == null) { |
- _firstComment = new StringToken(type, value, _tokenStart); |
- _lastComment = _firstComment; |
- } else { |
- _lastComment = _lastComment.setNext(new StringToken(type, value, _tokenStart)); |
- } |
- } |
- void appendEndToken(TokenType type32, TokenType beginType) { |
- Token token; |
- if (_firstComment == null) { |
- token = new Token(type32, _tokenStart); |
- } else { |
- token = new TokenWithComment(type32, _tokenStart, _firstComment); |
- _firstComment = null; |
- _lastComment = null; |
- } |
- _tail = _tail.setNext(token); |
- int last = _groupingStack.length - 1; |
- if (last >= 0) { |
- BeginToken begin = _groupingStack[last]; |
- if (identical(begin.type, beginType)) { |
- begin.endToken = token; |
- _groupingStack.removeAt(last); |
- } |
- } |
- } |
- void appendEofToken() { |
- Token eofToken; |
- if (_firstComment == null) { |
- eofToken = new Token(TokenType.EOF, offset + 1); |
- } else { |
- eofToken = new TokenWithComment(TokenType.EOF, offset + 1, _firstComment); |
- _firstComment = null; |
- _lastComment = null; |
- } |
- eofToken.setNext(eofToken); |
- _tail = _tail.setNext(eofToken); |
- if (!_groupingStack.isEmpty) { |
- _hasUnmatchedGroups2 = true; |
- } |
- } |
- void appendKeywordToken(Keyword keyword) { |
- if (_firstComment == null) { |
- _tail = _tail.setNext(new KeywordToken(keyword, _tokenStart)); |
- } else { |
- _tail = _tail.setNext(new KeywordTokenWithComment(keyword, _tokenStart, _firstComment)); |
- _firstComment = null; |
- _lastComment = null; |
- } |
- } |
- void appendStringToken(TokenType type, String value) { |
- if (_firstComment == null) { |
- _tail = _tail.setNext(new StringToken(type, value, _tokenStart)); |
- } else { |
- _tail = _tail.setNext(new StringTokenWithComment(type, value, _tokenStart, _firstComment)); |
- _firstComment = null; |
- _lastComment = null; |
- } |
- } |
- void appendStringToken2(TokenType type, String value, int offset) { |
- if (_firstComment == null) { |
- _tail = _tail.setNext(new StringToken(type, value, _tokenStart + offset)); |
- } else { |
- _tail = _tail.setNext(new StringTokenWithComment(type, value, _tokenStart + offset, _firstComment)); |
- _firstComment = null; |
- _lastComment = null; |
- } |
- } |
- void appendToken(TokenType type) { |
- if (_firstComment == null) { |
- _tail = _tail.setNext(new Token(type, _tokenStart)); |
- } else { |
- _tail = _tail.setNext(new TokenWithComment(type, _tokenStart, _firstComment)); |
- _firstComment = null; |
- _lastComment = null; |
- } |
- } |
- void appendToken2(TokenType type, int offset) { |
- if (_firstComment == null) { |
- _tail = _tail.setNext(new Token(type, offset)); |
- } else { |
- _tail = _tail.setNext(new TokenWithComment(type, offset, _firstComment)); |
- _firstComment = null; |
- _lastComment = null; |
- } |
- } |
- void beginToken() { |
- _tokenStart = offset; |
- } |
- int bigSwitch(int next) { |
- beginToken(); |
- if (next == 0xD) { |
- next = advance(); |
- if (next == 0xA) { |
- next = advance(); |
- } |
- recordStartOfLine(); |
- return next; |
- } else if (next == 0xA) { |
- recordStartOfLine(); |
- return advance(); |
- } else if (next == 0x9 || next == 0x20) { |
- return advance(); |
- } |
- if (next == 0x72) { |
- int peek3 = peek(); |
- if (peek3 == 0x22 || peek3 == 0x27) { |
- int start = offset; |
- return tokenizeString(advance(), start, true); |
- } |
- } |
- if (0x61 <= next && next <= 0x7A) { |
- return tokenizeKeywordOrIdentifier(next, true); |
- } |
- if ((0x41 <= next && next <= 0x5A) || next == 0x5F || next == 0x24) { |
- return tokenizeIdentifier(next, offset, true); |
- } |
- if (next == 0x3C) { |
- return tokenizeLessThan(next); |
- } |
- if (next == 0x3E) { |
- return tokenizeGreaterThan(next); |
- } |
- if (next == 0x3D) { |
- return tokenizeEquals(next); |
- } |
- if (next == 0x21) { |
- return tokenizeExclamation(next); |
- } |
- if (next == 0x2B) { |
- return tokenizePlus(next); |
- } |
- if (next == 0x2D) { |
- return tokenizeMinus(next); |
- } |
- if (next == 0x2A) { |
- return tokenizeMultiply(next); |
- } |
- if (next == 0x25) { |
- return tokenizePercent(next); |
- } |
- if (next == 0x26) { |
- return tokenizeAmpersand(next); |
- } |
- if (next == 0x7C) { |
- return tokenizeBar(next); |
- } |
- if (next == 0x5E) { |
- return tokenizeCaret(next); |
- } |
- if (next == 0x5B) { |
- return tokenizeOpenSquareBracket(next); |
- } |
- if (next == 0x7E) { |
- return tokenizeTilde(next); |
- } |
- if (next == 0x5C) { |
- appendToken(TokenType.BACKSLASH); |
- return advance(); |
- } |
- if (next == 0x23) { |
- return tokenizeTag(next); |
- } |
- if (next == 0x28) { |
- appendBeginToken(TokenType.OPEN_PAREN); |
- return advance(); |
- } |
- if (next == 0x29) { |
- appendEndToken(TokenType.CLOSE_PAREN, TokenType.OPEN_PAREN); |
- return advance(); |
- } |
- if (next == 0x2C) { |
- appendToken(TokenType.COMMA); |
- return advance(); |
- } |
- if (next == 0x3A) { |
- appendToken(TokenType.COLON); |
- return advance(); |
- } |
- if (next == 0x3B) { |
- appendToken(TokenType.SEMICOLON); |
- return advance(); |
- } |
- if (next == 0x3F) { |
- appendToken(TokenType.QUESTION); |
- return advance(); |
- } |
- if (next == 0x5D) { |
- appendEndToken(TokenType.CLOSE_SQUARE_BRACKET, TokenType.OPEN_SQUARE_BRACKET); |
- return advance(); |
- } |
- if (next == 0x60) { |
- appendToken(TokenType.BACKPING); |
- return advance(); |
- } |
- if (next == 0x7B) { |
- appendBeginToken(TokenType.OPEN_CURLY_BRACKET); |
- return advance(); |
- } |
- if (next == 0x7D) { |
- appendEndToken(TokenType.CLOSE_CURLY_BRACKET, TokenType.OPEN_CURLY_BRACKET); |
- return advance(); |
- } |
- if (next == 0x2F) { |
- return tokenizeSlashOrComment(next); |
- } |
- if (next == 0x40) { |
- appendToken(TokenType.AT); |
- return advance(); |
- } |
- if (next == 0x22 || next == 0x27) { |
- return tokenizeString(next, offset, false); |
- } |
- if (next == 0x2E) { |
- return tokenizeDotOrNumber(next); |
- } |
- if (next == 0x30) { |
- return tokenizeHexOrNumber(next); |
- } |
- if (0x31 <= next && next <= 0x39) { |
- return tokenizeNumber(next); |
- } |
- if (next == -1) { |
- return -1; |
- } |
- if (Character.isLetter(next)) { |
- return tokenizeIdentifier(next, offset, true); |
- } |
- if (next == _$NBSP) { |
- return advance(); |
- } |
- reportError(ScannerErrorCode.ILLEGAL_CHARACTER, [next]); |
- return advance(); |
- } |
- /** |
- * Return the beginning token corresponding to a closing brace that was found while scanning |
- * inside a string interpolation expression. Tokens that cannot be matched with the closing brace |
- * will be dropped from the stack. |
- * @return the token to be paired with the closing brace |
- */ |
- BeginToken findTokenMatchingClosingBraceInInterpolationExpression() { |
- int last = _groupingStack.length - 1; |
- while (last >= 0) { |
- BeginToken begin = _groupingStack[last]; |
- if (identical(begin.type, TokenType.OPEN_CURLY_BRACKET) || identical(begin.type, TokenType.STRING_INTERPOLATION_EXPRESSION)) { |
- return begin; |
- } |
- _hasUnmatchedGroups2 = true; |
- _groupingStack.removeAt(last); |
- last--; |
- } |
- return null; |
- } |
- Token firstToken() => _tokens.next; |
- /** |
- * Return the source being scanned. |
- * @return the source being scanned |
- */ |
- Source get source => _source; |
- /** |
- * Report an error at the current offset. |
- * @param errorCode the error code indicating the nature of the error |
- * @param arguments any arguments needed to complete the error message |
- */ |
- void reportError(ScannerErrorCode errorCode, List<Object> arguments) { |
- _errorListener.onError(new AnalysisError.con2(source, offset, 1, errorCode, [arguments])); |
- } |
- int select(int choice, TokenType yesType, TokenType noType) { |
- int next = advance(); |
- if (next == choice) { |
- appendToken(yesType); |
- return advance(); |
- } else { |
- appendToken(noType); |
- return next; |
- } |
- } |
- int select2(int choice, TokenType yesType, TokenType noType, int offset) { |
- int next = advance(); |
- if (next == choice) { |
- appendToken2(yesType, offset); |
- return advance(); |
- } else { |
- appendToken2(noType, offset); |
- return next; |
- } |
- } |
- int tokenizeAmpersand(int next) { |
- next = advance(); |
- if (next == 0x26) { |
- appendToken(TokenType.AMPERSAND_AMPERSAND); |
- return advance(); |
- } else if (next == 0x3D) { |
- appendToken(TokenType.AMPERSAND_EQ); |
- return advance(); |
- } else { |
- appendToken(TokenType.AMPERSAND); |
- return next; |
- } |
- } |
- int tokenizeBar(int next) { |
- next = advance(); |
- if (next == 0x7C) { |
- appendToken(TokenType.BAR_BAR); |
- return advance(); |
- } else if (next == 0x3D) { |
- appendToken(TokenType.BAR_EQ); |
- return advance(); |
- } else { |
- appendToken(TokenType.BAR); |
- return next; |
- } |
- } |
- int tokenizeCaret(int next) => select(0x3D, TokenType.CARET_EQ, TokenType.CARET); |
- int tokenizeDotOrNumber(int next) { |
- int start = offset; |
- next = advance(); |
- if ((0x30 <= next && next <= 0x39)) { |
- return tokenizeFractionPart(next, start); |
- } else if (0x2E == next) { |
- return select(0x2E, TokenType.PERIOD_PERIOD_PERIOD, TokenType.PERIOD_PERIOD); |
- } else { |
- appendToken(TokenType.PERIOD); |
- return next; |
- } |
- } |
- int tokenizeEquals(int next) { |
- next = advance(); |
- if (next == 0x3D) { |
- appendToken(TokenType.EQ_EQ); |
- return advance(); |
- } else if (next == 0x3E) { |
- appendToken(TokenType.FUNCTION); |
- return advance(); |
- } |
- appendToken(TokenType.EQ); |
- return next; |
- } |
- int tokenizeExclamation(int next) { |
- next = advance(); |
- if (next == 0x3D) { |
- appendToken(TokenType.BANG_EQ); |
- return advance(); |
- } |
- appendToken(TokenType.BANG); |
- return next; |
- } |
- int tokenizeExponent(int next) { |
- if (next == 0x2B || next == 0x2D) { |
- next = advance(); |
- } |
- bool hasDigits = false; |
- while (true) { |
- if (0x30 <= next && next <= 0x39) { |
- hasDigits = true; |
- } else { |
- if (!hasDigits) { |
- reportError(ScannerErrorCode.MISSING_DIGIT, []); |
- } |
- return next; |
- } |
- next = advance(); |
- } |
- } |
- int tokenizeFractionPart(int next, int start) { |
- bool done = false; |
- bool hasDigit = false; |
- LOOP: while (!done) { |
- if (0x30 <= next && next <= 0x39) { |
- hasDigit = true; |
- } else if (0x65 == next || 0x45 == next) { |
- hasDigit = true; |
- next = tokenizeExponent(advance()); |
- done = true; |
- continue LOOP; |
- } else { |
- done = true; |
- continue LOOP; |
- } |
- next = advance(); |
- } |
- if (!hasDigit) { |
- appendStringToken(TokenType.INT, getString(start, -2)); |
- if (0x2E == next) { |
- return select2(0x2E, TokenType.PERIOD_PERIOD_PERIOD, TokenType.PERIOD_PERIOD, offset - 1); |
- } |
- appendToken2(TokenType.PERIOD, offset - 1); |
- return bigSwitch(next); |
- } |
- if (next == 0x64 || next == 0x44) { |
- next = advance(); |
- } |
- appendStringToken(TokenType.DOUBLE, getString(start, next < 0 ? 0 : -1)); |
- return next; |
- } |
- int tokenizeGreaterThan(int next) { |
- next = advance(); |
- if (0x3D == next) { |
- appendToken(TokenType.GT_EQ); |
- return advance(); |
- } else if (0x3E == next) { |
- next = advance(); |
- if (0x3D == next) { |
- appendToken(TokenType.GT_GT_EQ); |
- return advance(); |
- } else { |
- appendToken(TokenType.GT_GT); |
- return next; |
- } |
- } else { |
- appendToken(TokenType.GT); |
- return next; |
- } |
- } |
- int tokenizeHex(int next) { |
- int start = offset - 1; |
- bool hasDigits = false; |
- while (true) { |
- next = advance(); |
- if ((0x30 <= next && next <= 0x39) || (0x41 <= next && next <= 0x46) || (0x61 <= next && next <= 0x66)) { |
- hasDigits = true; |
- } else { |
- if (!hasDigits) { |
- reportError(ScannerErrorCode.MISSING_HEX_DIGIT, []); |
- } |
- appendStringToken(TokenType.HEXADECIMAL, getString(start, next < 0 ? 0 : -1)); |
- return next; |
- } |
- } |
- } |
- int tokenizeHexOrNumber(int next) { |
- int x = peek(); |
- if (x == 0x78 || x == 0x58) { |
- advance(); |
- return tokenizeHex(x); |
- } |
- return tokenizeNumber(next); |
- } |
- int tokenizeIdentifier(int next, int start, bool allowDollar) { |
- while ((0x61 <= next && next <= 0x7A) || (0x41 <= next && next <= 0x5A) || (0x30 <= next && next <= 0x39) || next == 0x5F || (next == 0x24 && allowDollar) || Character.isLetterOrDigit(next)) { |
- next = advance(); |
- } |
- appendStringToken(TokenType.IDENTIFIER, getString(start, next < 0 ? 0 : -1)); |
- return next; |
- } |
- int tokenizeInterpolatedExpression(int next, int start) { |
- appendBeginToken(TokenType.STRING_INTERPOLATION_EXPRESSION); |
- next = advance(); |
- while (next != -1) { |
- if (next == 0x7D) { |
- BeginToken begin = findTokenMatchingClosingBraceInInterpolationExpression(); |
- if (begin == null) { |
- beginToken(); |
- appendToken(TokenType.CLOSE_CURLY_BRACKET); |
- next = advance(); |
- beginToken(); |
- return next; |
- } else if (identical(begin.type, TokenType.OPEN_CURLY_BRACKET)) { |
- beginToken(); |
- appendEndToken(TokenType.CLOSE_CURLY_BRACKET, TokenType.OPEN_CURLY_BRACKET); |
- next = advance(); |
- beginToken(); |
- } else if (identical(begin.type, TokenType.STRING_INTERPOLATION_EXPRESSION)) { |
- beginToken(); |
- appendEndToken(TokenType.CLOSE_CURLY_BRACKET, TokenType.STRING_INTERPOLATION_EXPRESSION); |
- next = advance(); |
- beginToken(); |
- return next; |
- } |
- } else { |
- next = bigSwitch(next); |
- } |
- } |
- if (next == -1) { |
- return next; |
- } |
- next = advance(); |
- beginToken(); |
- return next; |
- } |
- int tokenizeInterpolatedIdentifier(int next, int start) { |
- appendStringToken2(TokenType.STRING_INTERPOLATION_IDENTIFIER, "\$", 0); |
- beginToken(); |
- next = tokenizeKeywordOrIdentifier(next, false); |
- beginToken(); |
- return next; |
- } |
- int tokenizeKeywordOrIdentifier(int next2, bool allowDollar) { |
- KeywordState state = KeywordState.KEYWORD_STATE; |
- int start = offset; |
- while (state != null && 0x61 <= next2 && next2 <= 0x7A) { |
- state = state.next((next2 as int)); |
- next2 = advance(); |
- } |
- if (state == null || state.keyword() == null) { |
- return tokenizeIdentifier(next2, start, allowDollar); |
- } |
- if ((0x41 <= next2 && next2 <= 0x5A) || (0x30 <= next2 && next2 <= 0x39) || next2 == 0x5F || next2 == 0x24) { |
- return tokenizeIdentifier(next2, start, allowDollar); |
- } else if (next2 < 128) { |
- appendKeywordToken(state.keyword()); |
- return next2; |
- } else { |
- return tokenizeIdentifier(next2, start, allowDollar); |
- } |
- } |
- int tokenizeLessThan(int next) { |
- next = advance(); |
- if (0x3D == next) { |
- appendToken(TokenType.LT_EQ); |
- return advance(); |
- } else if (0x3C == next) { |
- return select(0x3D, TokenType.LT_LT_EQ, TokenType.LT_LT); |
- } else { |
- appendToken(TokenType.LT); |
- return next; |
- } |
- } |
- int tokenizeMinus(int next) { |
- next = advance(); |
- if (next == 0x2D) { |
- appendToken(TokenType.MINUS_MINUS); |
- return advance(); |
- } else if (next == 0x3D) { |
- appendToken(TokenType.MINUS_EQ); |
- return advance(); |
- } else { |
- appendToken(TokenType.MINUS); |
- return next; |
- } |
- } |
- int tokenizeMultiLineComment(int next) { |
- int nesting = 1; |
- next = advance(); |
- while (true) { |
- if (-1 == next) { |
- reportError(ScannerErrorCode.UNTERMINATED_MULTI_LINE_COMMENT, []); |
- appendCommentToken(TokenType.MULTI_LINE_COMMENT, getString(_tokenStart, 0)); |
- return next; |
- } else if (0x2A == next) { |
- next = advance(); |
- if (0x2F == next) { |
- --nesting; |
- if (0 == nesting) { |
- appendCommentToken(TokenType.MULTI_LINE_COMMENT, getString(_tokenStart, 0)); |
- return advance(); |
- } else { |
- next = advance(); |
- } |
- } |
- } else if (0x2F == next) { |
- next = advance(); |
- if (0x2A == next) { |
- next = advance(); |
- ++nesting; |
- } |
- } else { |
- next = advance(); |
- } |
- } |
- } |
- int tokenizeMultiLineRawString(int quoteChar, int start) { |
- int next = advance(); |
- outer: while (next != -1) { |
- while (next != quoteChar) { |
- next = advance(); |
- if (next == -1) { |
- break outer; |
- } |
- } |
- next = advance(); |
- if (next == quoteChar) { |
- next = advance(); |
- if (next == quoteChar) { |
- appendStringToken(TokenType.STRING, getString(start, 0)); |
- return advance(); |
- } |
- } |
- } |
- reportError(ScannerErrorCode.UNTERMINATED_STRING_LITERAL, []); |
- appendStringToken(TokenType.STRING, getString(start, 0)); |
- return advance(); |
- } |
- int tokenizeMultiLineString(int quoteChar, int start, bool raw) { |
- if (raw) { |
- return tokenizeMultiLineRawString(quoteChar, start); |
- } |
- int next = advance(); |
- while (next != -1) { |
- if (next == 0x24) { |
- appendStringToken(TokenType.STRING, getString(start, -1)); |
- beginToken(); |
- next = tokenizeStringInterpolation(start); |
- start = offset; |
- continue; |
- } |
- if (next == quoteChar) { |
- next = advance(); |
- if (next == quoteChar) { |
- next = advance(); |
- if (next == quoteChar) { |
- appendStringToken(TokenType.STRING, getString(start, 0)); |
- return advance(); |
- } |
- } |
- continue; |
- } |
- if (next == 0x5C) { |
- next = advance(); |
- if (next == -1) { |
- break; |
- } |
- } |
- next = advance(); |
- } |
- reportError(ScannerErrorCode.UNTERMINATED_STRING_LITERAL, []); |
- appendStringToken(TokenType.STRING, getString(start, 0)); |
- return advance(); |
- } |
- int tokenizeMultiply(int next) => select(0x3D, TokenType.STAR_EQ, TokenType.STAR); |
- int tokenizeNumber(int next) { |
- int start = offset; |
- while (true) { |
- next = advance(); |
- if (0x30 <= next && next <= 0x39) { |
- continue; |
- } else if (next == 0x2E) { |
- return tokenizeFractionPart(advance(), start); |
- } else if (next == 0x64 || next == 0x44) { |
- appendStringToken(TokenType.DOUBLE, getString(start, 0)); |
- return advance(); |
- } else if (next == 0x65 || next == 0x45) { |
- return tokenizeFractionPart(next, start); |
- } else { |
- appendStringToken(TokenType.INT, getString(start, next < 0 ? 0 : -1)); |
- return next; |
- } |
- } |
- } |
- int tokenizeOpenSquareBracket(int next) { |
- next = advance(); |
- if (next == 0x5D) { |
- return select(0x3D, TokenType.INDEX_EQ, TokenType.INDEX); |
- } else { |
- appendBeginToken(TokenType.OPEN_SQUARE_BRACKET); |
- return next; |
- } |
- } |
- int tokenizePercent(int next) => select(0x3D, TokenType.PERCENT_EQ, TokenType.PERCENT); |
- int tokenizePlus(int next) { |
- next = advance(); |
- if (0x2B == next) { |
- appendToken(TokenType.PLUS_PLUS); |
- return advance(); |
- } else if (0x3D == next) { |
- appendToken(TokenType.PLUS_EQ); |
- return advance(); |
- } else { |
- appendToken(TokenType.PLUS); |
- return next; |
- } |
- } |
- int tokenizeSingleLineComment(int next) { |
- while (true) { |
- next = advance(); |
- if (0xA == next || 0xD == next || -1 == next) { |
- appendCommentToken(TokenType.SINGLE_LINE_COMMENT, getString(_tokenStart, 0)); |
- return next; |
- } |
- } |
- } |
- int tokenizeSingleLineRawString(int next, int quoteChar, int start) { |
- next = advance(); |
- while (next != -1) { |
- if (next == quoteChar) { |
- appendStringToken(TokenType.STRING, getString(start, 0)); |
- return advance(); |
- } else if (next == 0xD || next == 0xA) { |
- reportError(ScannerErrorCode.UNTERMINATED_STRING_LITERAL, []); |
- appendStringToken(TokenType.STRING, getString(start, 0)); |
- return advance(); |
- } |
- next = advance(); |
- } |
- reportError(ScannerErrorCode.UNTERMINATED_STRING_LITERAL, []); |
- appendStringToken(TokenType.STRING, getString(start, 0)); |
- return advance(); |
- } |
- int tokenizeSingleLineString(int next, int quoteChar, int start) { |
- while (next != quoteChar) { |
- if (next == 0x5C) { |
- next = advance(); |
- } else if (next == 0x24) { |
- appendStringToken(TokenType.STRING, getString(start, -1)); |
- beginToken(); |
- next = tokenizeStringInterpolation(start); |
- start = offset; |
- continue; |
- } |
- if (next <= 0xD && (next == 0xA || next == 0xD || next == -1)) { |
- reportError(ScannerErrorCode.UNTERMINATED_STRING_LITERAL, []); |
- appendStringToken(TokenType.STRING, getString(start, 0)); |
- return advance(); |
- } |
- next = advance(); |
- } |
- appendStringToken(TokenType.STRING, getString(start, 0)); |
- return advance(); |
- } |
- int tokenizeSlashOrComment(int next) { |
- next = advance(); |
- if (0x2A == next) { |
- return tokenizeMultiLineComment(next); |
- } else if (0x2F == next) { |
- return tokenizeSingleLineComment(next); |
- } else if (0x3D == next) { |
- appendToken(TokenType.SLASH_EQ); |
- return advance(); |
- } else { |
- appendToken(TokenType.SLASH); |
- return next; |
- } |
- } |
- int tokenizeString(int next, int start, bool raw) { |
- int quoteChar = next; |
- next = advance(); |
- if (quoteChar == next) { |
- next = advance(); |
- if (quoteChar == next) { |
- return tokenizeMultiLineString(quoteChar, start, raw); |
- } else { |
- appendStringToken(TokenType.STRING, getString(start, -1)); |
- return next; |
- } |
- } |
- if (raw) { |
- return tokenizeSingleLineRawString(next, quoteChar, start); |
- } else { |
- return tokenizeSingleLineString(next, quoteChar, start); |
- } |
- } |
- int tokenizeStringInterpolation(int start) { |
- beginToken(); |
- int next = advance(); |
- if (next == 0x7B) { |
- return tokenizeInterpolatedExpression(next, start); |
- } else { |
- return tokenizeInterpolatedIdentifier(next, start); |
- } |
- } |
- int tokenizeTag(int next) { |
- if (offset == 0) { |
- if (peek() == 0x21) { |
- do { |
- next = advance(); |
- } while (next != 0xA && next != 0xD && next > 0); |
- appendStringToken(TokenType.SCRIPT_TAG, getString(_tokenStart, 0)); |
- return next; |
- } |
- } |
- appendToken(TokenType.HASH); |
- return advance(); |
- } |
- int tokenizeTilde(int next) { |
- next = advance(); |
- if (next == 0x2F) { |
- return select(0x3D, TokenType.TILDE_SLASH_EQ, TokenType.TILDE_SLASH); |
- } else { |
- appendToken(TokenType.TILDE); |
- return next; |
- } |
- } |
-} |
-/** |
- * Instances of the class {@code StringToken} represent a token whose value is independent of it's |
- * type. |
- * @coverage dart.engine.parser |
- */ |
-class StringToken extends Token { |
- /** |
- * The lexeme represented by this token. |
- */ |
- String _value2; |
- /** |
- * Initialize a newly created token to represent a token of the given type with the given value. |
- * @param type the type of the token |
- * @param value the lexeme represented by this token |
- * @param offset the offset from the beginning of the file to the first character in the token |
- */ |
- StringToken(TokenType type, String value, int offset) : super(type, offset) { |
- this._value2 = value; |
- } |
- String get lexeme => _value2; |
- String value() => _value2; |
-} |
-/** |
- * Instances of the class {@code CharBufferScanner} implement a scanner that reads from a character |
- * buffer. The scanning logic is in the superclass. |
- * @coverage dart.engine.parser |
- */ |
-class CharBufferScanner extends AbstractScanner { |
- /** |
- * The buffer from which characters will be read. |
- */ |
- CharBuffer _buffer; |
- /** |
- * The number of characters in the buffer. |
- */ |
- int _bufferLength = 0; |
- /** |
- * The index of the last character that was read. |
- */ |
- int _charOffset = 0; |
- /** |
- * Initialize a newly created scanner to scan the characters in the given character buffer. |
- * @param source the source being scanned |
- * @param buffer the buffer from which characters will be read |
- * @param errorListener the error listener that will be informed of any errors that are found |
- */ |
- CharBufferScanner(Source source, CharBuffer buffer, AnalysisErrorListener errorListener) : super(source, errorListener) { |
- this._buffer = buffer; |
- this._bufferLength = buffer.length(); |
- this._charOffset = -1; |
- } |
- int get offset => _charOffset; |
- int advance() { |
- if (_charOffset + 1 >= _bufferLength) { |
- return -1; |
- } |
- return _buffer.charAt(++_charOffset); |
- } |
- String getString(int start, int endDelta) => _buffer.subSequence(start, _charOffset + 1 + endDelta).toString(); |
- int peek() { |
- if (_charOffset + 1 >= _buffer.length()) { |
- return -1; |
- } |
- return _buffer.charAt(_charOffset + 1); |
- } |
-} |
-/** |
- * Instances of the class {@code TokenWithComment} represent a normal token that is preceded by |
- * comments. |
- * @coverage dart.engine.parser |
- */ |
-class TokenWithComment extends Token { |
- /** |
- * The first comment in the list of comments that precede this token. |
- */ |
- Token _precedingComment; |
- /** |
- * Initialize a newly created token to have the given type and offset and to be preceded by the |
- * comments reachable from the given comment. |
- * @param type the type of the token |
- * @param offset the offset from the beginning of the file to the first character in the token |
- * @param precedingComment the first comment in the list of comments that precede this token |
- */ |
- TokenWithComment(TokenType type, int offset, Token precedingComment) : super(type, offset) { |
- this._precedingComment = precedingComment; |
- } |
- Token get precedingComments => _precedingComment; |
-} |
-/** |
- * Instances of the class {@code Token} represent a token that was scanned from the input. Each |
- * token knows which token follows it, acting as the head of a linked list of tokens. |
- * @coverage dart.engine.parser |
- */ |
-class Token { |
- /** |
- * The type of the token. |
- */ |
- TokenType _type; |
- /** |
- * The offset from the beginning of the file to the first character in the token. |
- */ |
- int _offset = 0; |
- /** |
- * The previous token in the token stream. |
- */ |
- Token _previous; |
- /** |
- * The next token in the token stream. |
- */ |
- Token _next; |
- /** |
- * Initialize a newly created token to have the given type and offset. |
- * @param type the type of the token |
- * @param offset the offset from the beginning of the file to the first character in the token |
- */ |
- Token(TokenType type, int offset) { |
- this._type = type; |
- this._offset = offset; |
- } |
- /** |
- * Return the offset from the beginning of the file to the character after last character of the |
- * token. |
- * @return the offset from the beginning of the file to the first character after last character |
- * of the token |
- */ |
- int get end => _offset + length; |
- /** |
- * Return the number of characters in the node's source range. |
- * @return the number of characters in the node's source range |
- */ |
- int get length => lexeme.length; |
- /** |
- * Return the lexeme that represents this token. |
- * @return the lexeme that represents this token |
- */ |
- String get lexeme => _type.lexeme; |
- /** |
- * Return the next token in the token stream. |
- * @return the next token in the token stream |
- */ |
- Token get next => _next; |
- /** |
- * Return the offset from the beginning of the file to the first character in the token. |
- * @return the offset from the beginning of the file to the first character in the token |
- */ |
- int get offset => _offset; |
- /** |
- * Return the first comment in the list of comments that precede this token, or {@code null} if |
- * there are no comments preceding this token. Additional comments can be reached by following the |
- * token stream using {@link #getNext()} until {@code null} is returned. |
- * @return the first comment in the list of comments that precede this token |
- */ |
- Token get precedingComments => null; |
- /** |
- * Return the previous token in the token stream. |
- * @return the previous token in the token stream |
- */ |
- Token get previous => _previous; |
- /** |
- * Return the type of the token. |
- * @return the type of the token |
- */ |
- TokenType get type => _type; |
- /** |
- * Return {@code true} if this token represents an operator. |
- * @return {@code true} if this token represents an operator |
- */ |
- bool isOperator() => _type.isOperator(); |
- /** |
- * Return {@code true} if this token is a synthetic token. A synthetic token is a token that was |
- * introduced by the parser in order to recover from an error in the code. Synthetic tokens always |
- * have a length of zero ({@code 0}). |
- * @return {@code true} if this token is a synthetic token |
- */ |
- bool isSynthetic() => length == 0; |
- /** |
- * Return {@code true} if this token represents an operator that can be defined by users. |
- * @return {@code true} if this token represents an operator that can be defined by users |
- */ |
- bool isUserDefinableOperator() => _type.isUserDefinableOperator(); |
- /** |
- * Set the next token in the token stream to the given token. This has the side-effect of setting |
- * this token to be the previous token for the given token. |
- * @param token the next token in the token stream |
- * @return the token that was passed in |
- */ |
- Token setNext(Token token) { |
- _next = token; |
- token.previous = this; |
- return token; |
- } |
- /** |
- * Set the next token in the token stream to the given token without changing which token is the |
- * previous token for the given token. |
- * @param token the next token in the token stream |
- * @return the token that was passed in |
- */ |
- Token setNextWithoutSettingPrevious(Token token) { |
- _next = token; |
- return token; |
- } |
- /** |
- * Set the offset from the beginning of the file to the first character in the token to the given |
- * offset. |
- * @param offset the offset from the beginning of the file to the first character in the token |
- */ |
- void set offset(int offset4) { |
- this._offset = offset4; |
- } |
- String toString() => lexeme; |
- /** |
- * Return the value of this token. For keyword tokens, this is the keyword associated with the |
- * token, for other tokens it is the lexeme associated with the token. |
- * @return the value of this token |
- */ |
- Object value() => _type.lexeme; |
- /** |
- * Set the previous token in the token stream to the given token. |
- * @param previous the previous token in the token stream |
- */ |
- void set previous(Token previous3) { |
- this._previous = previous3; |
- } |
-} |
-/** |
- * Instances of the class {@code StringScanner} implement a scanner that reads from a string. The |
- * scanning logic is in the superclass. |
- * @coverage dart.engine.parser |
- */ |
-class StringScanner extends AbstractScanner { |
- /** |
- * The offset from the beginning of the file to the beginning of the source being scanned. |
- */ |
- int _offsetDelta = 0; |
- /** |
- * The string from which characters will be read. |
- */ |
- String _string; |
- /** |
- * The number of characters in the string. |
- */ |
- int _stringLength = 0; |
- /** |
- * The index, relative to the string, of the last character that was read. |
- */ |
- int _charOffset = 0; |
- /** |
- * Initialize a newly created scanner to scan the characters in the given string. |
- * @param source the source being scanned |
- * @param string the string from which characters will be read |
- * @param errorListener the error listener that will be informed of any errors that are found |
- */ |
- StringScanner(Source source, String string, AnalysisErrorListener errorListener) : super(source, errorListener) { |
- this._offsetDelta = 0; |
- this._string = string; |
- this._stringLength = string.length; |
- this._charOffset = -1; |
- } |
- int get offset => _offsetDelta + _charOffset; |
- /** |
- * Record that the source begins on the given line and column at the given offset. The line starts |
- * for lines before the given line will not be correct. |
- * <p> |
- * This method must be invoked at most one time and must be invoked before scanning begins. The |
- * values provided must be sensible. The results are undefined if these conditions are violated. |
- * @param line the one-based index of the line containing the first character of the source |
- * @param column the one-based index of the column in which the first character of the source |
- * occurs |
- * @param offset the zero-based offset from the beginning of the larger context to the first |
- * character of the source |
- */ |
- void setSourceStart(int line, int column, int offset) { |
- if (line < 1 || column < 1 || offset < 0 || (line + column - 2) >= offset) { |
- return; |
- } |
- _offsetDelta = 1; |
- for (int i = 2; i < line; i++) { |
- recordStartOfLine(); |
- } |
- _offsetDelta = offset - column + 1; |
- recordStartOfLine(); |
- _offsetDelta = offset; |
- } |
- int advance() { |
- if (_charOffset + 1 >= _stringLength) { |
- return -1; |
- } |
- return _string.codeUnitAt(++_charOffset); |
- } |
- String getString(int start, int endDelta) => _string.substring(start - _offsetDelta, _charOffset + 1 + endDelta); |
- int peek() { |
- if (_charOffset + 1 >= _string.length) { |
- return -1; |
- } |
- return _string.codeUnitAt(_charOffset + 1); |
- } |
-} |
-/** |
- * Instances of the class {@code BeginTokenWithComment} represent a begin token that is preceded by |
- * comments. |
- * @coverage dart.engine.parser |
- */ |
-class BeginTokenWithComment extends BeginToken { |
- /** |
- * The first comment in the list of comments that precede this token. |
- */ |
- Token _precedingComment; |
- /** |
- * Initialize a newly created token to have the given type and offset and to be preceded by the |
- * comments reachable from the given comment. |
- * @param type the type of the token |
- * @param offset the offset from the beginning of the file to the first character in the token |
- * @param precedingComment the first comment in the list of comments that precede this token |
- */ |
- BeginTokenWithComment(TokenType type, int offset, Token precedingComment) : super(type, offset) { |
- this._precedingComment = precedingComment; |
- } |
- Token get precedingComments => _precedingComment; |
-} |
-/** |
- * Instances of the class {@code KeywordToken} represent a keyword in the language. |
- * @coverage dart.engine.parser |
- */ |
-class KeywordToken extends Token { |
- /** |
- * The keyword being represented by this token. |
- */ |
- Keyword _keyword; |
- /** |
- * Initialize a newly created token to represent the given keyword. |
- * @param keyword the keyword being represented by this token |
- * @param offset the offset from the beginning of the file to the first character in the token |
- */ |
- KeywordToken(Keyword keyword, int offset) : super(TokenType.KEYWORD, offset) { |
- this._keyword = keyword; |
- } |
- /** |
- * Return the keyword being represented by this token. |
- * @return the keyword being represented by this token |
- */ |
- Keyword get keyword => _keyword; |
- String get lexeme => _keyword.syntax; |
- Keyword value() => _keyword; |
-} |
-/** |
- * Instances of the class {@code BeginToken} represent the opening half of a grouping pair of |
- * tokens. This is used for curly brackets ('{'), parentheses ('('), and square brackets ('['). |
- * @coverage dart.engine.parser |
- */ |
-class BeginToken extends Token { |
- /** |
- * The token that corresponds to this token. |
- */ |
- Token _endToken; |
- /** |
- * Initialize a newly created token representing the opening half of a grouping pair of tokens. |
- * @param type the type of the token |
- * @param offset the offset from the beginning of the file to the first character in the token |
- */ |
- BeginToken(TokenType type, int offset) : super(type, offset) { |
- assert((identical(type, TokenType.OPEN_CURLY_BRACKET) || identical(type, TokenType.OPEN_PAREN) || identical(type, TokenType.OPEN_SQUARE_BRACKET) || identical(type, TokenType.STRING_INTERPOLATION_EXPRESSION))); |
- } |
- /** |
- * Return the token that corresponds to this token. |
- * @return the token that corresponds to this token |
- */ |
- Token get endToken => _endToken; |
- /** |
- * Set the token that corresponds to this token to the given token. |
- * @param token the token that corresponds to this token |
- */ |
- void set endToken(Token token) { |
- this._endToken = token; |
- } |
-} |
-/** |
- * The enumeration {@code TokenClass} represents classes (or groups) of tokens with a similar use. |
- * @coverage dart.engine.parser |
- */ |
-class TokenClass { |
- /** |
- * A value used to indicate that the token type is not part of any specific class of token. |
- */ |
- static final TokenClass NO_CLASS = new TokenClass.con1('NO_CLASS', 0); |
- /** |
- * A value used to indicate that the token type is an additive operator. |
- */ |
- static final TokenClass ADDITIVE_OPERATOR = new TokenClass.con2('ADDITIVE_OPERATOR', 1, 12); |
- /** |
- * A value used to indicate that the token type is an assignment operator. |
- */ |
- static final TokenClass ASSIGNMENT_OPERATOR = new TokenClass.con2('ASSIGNMENT_OPERATOR', 2, 1); |
- /** |
- * A value used to indicate that the token type is a bitwise-and operator. |
- */ |
- static final TokenClass BITWISE_AND_OPERATOR = new TokenClass.con2('BITWISE_AND_OPERATOR', 3, 8); |
- /** |
- * A value used to indicate that the token type is a bitwise-or operator. |
- */ |
- static final TokenClass BITWISE_OR_OPERATOR = new TokenClass.con2('BITWISE_OR_OPERATOR', 4, 6); |
- /** |
- * A value used to indicate that the token type is a bitwise-xor operator. |
- */ |
- static final TokenClass BITWISE_XOR_OPERATOR = new TokenClass.con2('BITWISE_XOR_OPERATOR', 5, 7); |
- /** |
- * A value used to indicate that the token type is a cascade operator. |
- */ |
- static final TokenClass CASCADE_OPERATOR = new TokenClass.con2('CASCADE_OPERATOR', 6, 2); |
- /** |
- * A value used to indicate that the token type is a conditional operator. |
- */ |
- static final TokenClass CONDITIONAL_OPERATOR = new TokenClass.con2('CONDITIONAL_OPERATOR', 7, 3); |
- /** |
- * A value used to indicate that the token type is an equality operator. |
- */ |
- static final TokenClass EQUALITY_OPERATOR = new TokenClass.con2('EQUALITY_OPERATOR', 8, 9); |
- /** |
- * A value used to indicate that the token type is a logical-and operator. |
- */ |
- static final TokenClass LOGICAL_AND_OPERATOR = new TokenClass.con2('LOGICAL_AND_OPERATOR', 9, 5); |
- /** |
- * A value used to indicate that the token type is a logical-or operator. |
- */ |
- static final TokenClass LOGICAL_OR_OPERATOR = new TokenClass.con2('LOGICAL_OR_OPERATOR', 10, 4); |
- /** |
- * A value used to indicate that the token type is a multiplicative operator. |
- */ |
- static final TokenClass MULTIPLICATIVE_OPERATOR = new TokenClass.con2('MULTIPLICATIVE_OPERATOR', 11, 13); |
- /** |
- * A value used to indicate that the token type is a relational operator. |
- */ |
- static final TokenClass RELATIONAL_OPERATOR = new TokenClass.con2('RELATIONAL_OPERATOR', 12, 10); |
- /** |
- * A value used to indicate that the token type is a shift operator. |
- */ |
- static final TokenClass SHIFT_OPERATOR = new TokenClass.con2('SHIFT_OPERATOR', 13, 11); |
- /** |
- * A value used to indicate that the token type is a unary operator. |
- */ |
- static final TokenClass UNARY_POSTFIX_OPERATOR = new TokenClass.con2('UNARY_POSTFIX_OPERATOR', 14, 15); |
- /** |
- * A value used to indicate that the token type is a unary operator. |
- */ |
- static final TokenClass UNARY_PREFIX_OPERATOR = new TokenClass.con2('UNARY_PREFIX_OPERATOR', 15, 14); |
- static final List<TokenClass> values = [NO_CLASS, ADDITIVE_OPERATOR, ASSIGNMENT_OPERATOR, BITWISE_AND_OPERATOR, BITWISE_OR_OPERATOR, BITWISE_XOR_OPERATOR, CASCADE_OPERATOR, CONDITIONAL_OPERATOR, EQUALITY_OPERATOR, LOGICAL_AND_OPERATOR, LOGICAL_OR_OPERATOR, MULTIPLICATIVE_OPERATOR, RELATIONAL_OPERATOR, SHIFT_OPERATOR, UNARY_POSTFIX_OPERATOR, UNARY_PREFIX_OPERATOR]; |
- String __name; |
- int __ordinal = 0; |
- int get ordinal => __ordinal; |
- /** |
- * The precedence of tokens of this class, or {@code 0} if the such tokens do not represent an |
- * operator. |
- */ |
- int _precedence = 0; |
- TokenClass.con1(String ___name, int ___ordinal) { |
- _jtd_constructor_275_impl(___name, ___ordinal); |
- } |
- _jtd_constructor_275_impl(String ___name, int ___ordinal) { |
- _jtd_constructor_276_impl(___name, ___ordinal, 0); |
- } |
- TokenClass.con2(String ___name, int ___ordinal, int precedence2) { |
- _jtd_constructor_276_impl(___name, ___ordinal, precedence2); |
- } |
- _jtd_constructor_276_impl(String ___name, int ___ordinal, int precedence2) { |
- __name = ___name; |
- __ordinal = ___ordinal; |
- this._precedence = precedence2; |
- } |
- /** |
- * Return the precedence of tokens of this class, or {@code 0} if the such tokens do not represent |
- * an operator. |
- * @return the precedence of tokens of this class |
- */ |
- int get precedence => _precedence; |
- String toString() => __name; |
-} |
-/** |
- * Instances of the class {@code KeywordTokenWithComment} implement a keyword token that is preceded |
- * by comments. |
- * @coverage dart.engine.parser |
- */ |
-class KeywordTokenWithComment extends KeywordToken { |
- /** |
- * The first comment in the list of comments that precede this token. |
- */ |
- Token _precedingComment; |
- /** |
- * Initialize a newly created token to to represent the given keyword and to be preceded by the |
- * comments reachable from the given comment. |
- * @param keyword the keyword being represented by this token |
- * @param offset the offset from the beginning of the file to the first character in the token |
- * @param precedingComment the first comment in the list of comments that precede this token |
- */ |
- KeywordTokenWithComment(Keyword keyword, int offset, Token precedingComment) : super(keyword, offset) { |
- this._precedingComment = precedingComment; |
- } |
- Token get precedingComments => _precedingComment; |
-} |
-/** |
- * The enumeration {@code TokenType} defines the types of tokens that can be returned by the |
- * scanner. |
- * @coverage dart.engine.parser |
- */ |
-class TokenType { |
- /** |
- * The type of the token that marks the end of the input. |
- */ |
- static final TokenType EOF = new TokenType_EOF('EOF', 0, null, ""); |
- static final TokenType DOUBLE = new TokenType.con1('DOUBLE', 1); |
- static final TokenType HEXADECIMAL = new TokenType.con1('HEXADECIMAL', 2); |
- static final TokenType IDENTIFIER = new TokenType.con1('IDENTIFIER', 3); |
- static final TokenType INT = new TokenType.con1('INT', 4); |
- static final TokenType KEYWORD = new TokenType.con1('KEYWORD', 5); |
- static final TokenType MULTI_LINE_COMMENT = new TokenType.con1('MULTI_LINE_COMMENT', 6); |
- static final TokenType SCRIPT_TAG = new TokenType.con1('SCRIPT_TAG', 7); |
- static final TokenType SINGLE_LINE_COMMENT = new TokenType.con1('SINGLE_LINE_COMMENT', 8); |
- static final TokenType STRING = new TokenType.con1('STRING', 9); |
- static final TokenType AMPERSAND = new TokenType.con2('AMPERSAND', 10, TokenClass.BITWISE_AND_OPERATOR, "&"); |
- static final TokenType AMPERSAND_AMPERSAND = new TokenType.con2('AMPERSAND_AMPERSAND', 11, TokenClass.LOGICAL_AND_OPERATOR, "&&"); |
- static final TokenType AMPERSAND_EQ = new TokenType.con2('AMPERSAND_EQ', 12, TokenClass.ASSIGNMENT_OPERATOR, "&="); |
- static final TokenType AT = new TokenType.con2('AT', 13, null, "@"); |
- static final TokenType BANG = new TokenType.con2('BANG', 14, TokenClass.UNARY_PREFIX_OPERATOR, "!"); |
- static final TokenType BANG_EQ = new TokenType.con2('BANG_EQ', 15, TokenClass.EQUALITY_OPERATOR, "!="); |
- static final TokenType BAR = new TokenType.con2('BAR', 16, TokenClass.BITWISE_OR_OPERATOR, "|"); |
- static final TokenType BAR_BAR = new TokenType.con2('BAR_BAR', 17, TokenClass.LOGICAL_OR_OPERATOR, "||"); |
- static final TokenType BAR_EQ = new TokenType.con2('BAR_EQ', 18, TokenClass.ASSIGNMENT_OPERATOR, "|="); |
- static final TokenType COLON = new TokenType.con2('COLON', 19, null, ":"); |
- static final TokenType COMMA = new TokenType.con2('COMMA', 20, null, ","); |
- static final TokenType CARET = new TokenType.con2('CARET', 21, TokenClass.BITWISE_XOR_OPERATOR, "^"); |
- static final TokenType CARET_EQ = new TokenType.con2('CARET_EQ', 22, TokenClass.ASSIGNMENT_OPERATOR, "^="); |
- static final TokenType CLOSE_CURLY_BRACKET = new TokenType.con2('CLOSE_CURLY_BRACKET', 23, null, "}"); |
- static final TokenType CLOSE_PAREN = new TokenType.con2('CLOSE_PAREN', 24, null, ")"); |
- static final TokenType CLOSE_SQUARE_BRACKET = new TokenType.con2('CLOSE_SQUARE_BRACKET', 25, null, "]"); |
- static final TokenType EQ = new TokenType.con2('EQ', 26, TokenClass.ASSIGNMENT_OPERATOR, "="); |
- static final TokenType EQ_EQ = new TokenType.con2('EQ_EQ', 27, TokenClass.EQUALITY_OPERATOR, "=="); |
- static final TokenType FUNCTION = new TokenType.con2('FUNCTION', 28, null, "=>"); |
- static final TokenType GT = new TokenType.con2('GT', 29, TokenClass.RELATIONAL_OPERATOR, ">"); |
- static final TokenType GT_EQ = new TokenType.con2('GT_EQ', 30, TokenClass.RELATIONAL_OPERATOR, ">="); |
- static final TokenType GT_GT = new TokenType.con2('GT_GT', 31, TokenClass.SHIFT_OPERATOR, ">>"); |
- static final TokenType GT_GT_EQ = new TokenType.con2('GT_GT_EQ', 32, TokenClass.ASSIGNMENT_OPERATOR, ">>="); |
- static final TokenType HASH = new TokenType.con2('HASH', 33, null, "#"); |
- static final TokenType INDEX = new TokenType.con2('INDEX', 34, TokenClass.UNARY_POSTFIX_OPERATOR, "[]"); |
- static final TokenType INDEX_EQ = new TokenType.con2('INDEX_EQ', 35, TokenClass.UNARY_POSTFIX_OPERATOR, "[]="); |
- static final TokenType IS = new TokenType.con2('IS', 36, TokenClass.RELATIONAL_OPERATOR, "is"); |
- static final TokenType LT = new TokenType.con2('LT', 37, TokenClass.RELATIONAL_OPERATOR, "<"); |
- static final TokenType LT_EQ = new TokenType.con2('LT_EQ', 38, TokenClass.RELATIONAL_OPERATOR, "<="); |
- static final TokenType LT_LT = new TokenType.con2('LT_LT', 39, TokenClass.SHIFT_OPERATOR, "<<"); |
- static final TokenType LT_LT_EQ = new TokenType.con2('LT_LT_EQ', 40, TokenClass.ASSIGNMENT_OPERATOR, "<<="); |
- static final TokenType MINUS = new TokenType.con2('MINUS', 41, TokenClass.ADDITIVE_OPERATOR, "-"); |
- static final TokenType MINUS_EQ = new TokenType.con2('MINUS_EQ', 42, TokenClass.ASSIGNMENT_OPERATOR, "-="); |
- static final TokenType MINUS_MINUS = new TokenType.con2('MINUS_MINUS', 43, TokenClass.UNARY_PREFIX_OPERATOR, "--"); |
- static final TokenType OPEN_CURLY_BRACKET = new TokenType.con2('OPEN_CURLY_BRACKET', 44, null, "{"); |
- static final TokenType OPEN_PAREN = new TokenType.con2('OPEN_PAREN', 45, TokenClass.UNARY_POSTFIX_OPERATOR, "("); |
- static final TokenType OPEN_SQUARE_BRACKET = new TokenType.con2('OPEN_SQUARE_BRACKET', 46, TokenClass.UNARY_POSTFIX_OPERATOR, "["); |
- static final TokenType PERCENT = new TokenType.con2('PERCENT', 47, TokenClass.MULTIPLICATIVE_OPERATOR, "%"); |
- static final TokenType PERCENT_EQ = new TokenType.con2('PERCENT_EQ', 48, TokenClass.ASSIGNMENT_OPERATOR, "%="); |
- static final TokenType PERIOD = new TokenType.con2('PERIOD', 49, TokenClass.UNARY_POSTFIX_OPERATOR, "."); |
- static final TokenType PERIOD_PERIOD = new TokenType.con2('PERIOD_PERIOD', 50, TokenClass.CASCADE_OPERATOR, ".."); |
- static final TokenType PLUS = new TokenType.con2('PLUS', 51, TokenClass.ADDITIVE_OPERATOR, "+"); |
- static final TokenType PLUS_EQ = new TokenType.con2('PLUS_EQ', 52, TokenClass.ASSIGNMENT_OPERATOR, "+="); |
- static final TokenType PLUS_PLUS = new TokenType.con2('PLUS_PLUS', 53, TokenClass.UNARY_PREFIX_OPERATOR, "++"); |
- static final TokenType QUESTION = new TokenType.con2('QUESTION', 54, TokenClass.CONDITIONAL_OPERATOR, "?"); |
- static final TokenType SEMICOLON = new TokenType.con2('SEMICOLON', 55, null, ";"); |
- static final TokenType SLASH = new TokenType.con2('SLASH', 56, TokenClass.MULTIPLICATIVE_OPERATOR, "/"); |
- static final TokenType SLASH_EQ = new TokenType.con2('SLASH_EQ', 57, TokenClass.ASSIGNMENT_OPERATOR, "/="); |
- static final TokenType STAR = new TokenType.con2('STAR', 58, TokenClass.MULTIPLICATIVE_OPERATOR, "*"); |
- static final TokenType STAR_EQ = new TokenType.con2('STAR_EQ', 59, TokenClass.ASSIGNMENT_OPERATOR, "*="); |
- static final TokenType STRING_INTERPOLATION_EXPRESSION = new TokenType.con2('STRING_INTERPOLATION_EXPRESSION', 60, null, "\${"); |
- static final TokenType STRING_INTERPOLATION_IDENTIFIER = new TokenType.con2('STRING_INTERPOLATION_IDENTIFIER', 61, null, "\$"); |
- static final TokenType TILDE = new TokenType.con2('TILDE', 62, TokenClass.UNARY_PREFIX_OPERATOR, "~"); |
- static final TokenType TILDE_SLASH = new TokenType.con2('TILDE_SLASH', 63, TokenClass.MULTIPLICATIVE_OPERATOR, "~/"); |
- static final TokenType TILDE_SLASH_EQ = new TokenType.con2('TILDE_SLASH_EQ', 64, TokenClass.ASSIGNMENT_OPERATOR, "~/="); |
- static final TokenType BACKPING = new TokenType.con2('BACKPING', 65, null, "`"); |
- static final TokenType BACKSLASH = new TokenType.con2('BACKSLASH', 66, null, "\\"); |
- static final TokenType PERIOD_PERIOD_PERIOD = new TokenType.con2('PERIOD_PERIOD_PERIOD', 67, null, "..."); |
- static final List<TokenType> values = [EOF, DOUBLE, HEXADECIMAL, IDENTIFIER, INT, KEYWORD, MULTI_LINE_COMMENT, SCRIPT_TAG, SINGLE_LINE_COMMENT, STRING, AMPERSAND, AMPERSAND_AMPERSAND, AMPERSAND_EQ, AT, BANG, BANG_EQ, BAR, BAR_BAR, BAR_EQ, COLON, COMMA, CARET, CARET_EQ, CLOSE_CURLY_BRACKET, CLOSE_PAREN, CLOSE_SQUARE_BRACKET, EQ, EQ_EQ, FUNCTION, GT, GT_EQ, GT_GT, GT_GT_EQ, HASH, INDEX, INDEX_EQ, IS, LT, LT_EQ, LT_LT, LT_LT_EQ, MINUS, MINUS_EQ, MINUS_MINUS, OPEN_CURLY_BRACKET, OPEN_PAREN, OPEN_SQUARE_BRACKET, PERCENT, PERCENT_EQ, PERIOD, PERIOD_PERIOD, PLUS, PLUS_EQ, PLUS_PLUS, QUESTION, SEMICOLON, SLASH, SLASH_EQ, STAR, STAR_EQ, STRING_INTERPOLATION_EXPRESSION, STRING_INTERPOLATION_IDENTIFIER, TILDE, TILDE_SLASH, TILDE_SLASH_EQ, BACKPING, BACKSLASH, PERIOD_PERIOD_PERIOD]; |
- String __name; |
- int __ordinal = 0; |
- int get ordinal => __ordinal; |
- /** |
- * The class of the token. |
- */ |
- TokenClass _tokenClass; |
- /** |
- * The lexeme that defines this type of token, or {@code null} if there is more than one possible |
- * lexeme for this type of token. |
- */ |
- String _lexeme; |
- TokenType.con1(String ___name, int ___ordinal) { |
- _jtd_constructor_277_impl(___name, ___ordinal); |
- } |
- _jtd_constructor_277_impl(String ___name, int ___ordinal) { |
- _jtd_constructor_278_impl(___name, ___ordinal, TokenClass.NO_CLASS, null); |
- } |
- TokenType.con2(String ___name, int ___ordinal, TokenClass tokenClass2, String lexeme2) { |
- _jtd_constructor_278_impl(___name, ___ordinal, tokenClass2, lexeme2); |
- } |
- _jtd_constructor_278_impl(String ___name, int ___ordinal, TokenClass tokenClass2, String lexeme2) { |
- __name = ___name; |
- __ordinal = ___ordinal; |
- this._tokenClass = tokenClass2 == null ? TokenClass.NO_CLASS : tokenClass2; |
- this._lexeme = lexeme2; |
- } |
- /** |
- * Return the lexeme that defines this type of token, or {@code null} if there is more than one |
- * possible lexeme for this type of token. |
- * @return the lexeme that defines this type of token |
- */ |
- String get lexeme => _lexeme; |
- /** |
- * Return the precedence of the token, or {@code 0} if the token does not represent an operator. |
- * @return the precedence of the token |
- */ |
- int get precedence => _tokenClass.precedence; |
- /** |
- * Return {@code true} if this type of token represents an additive operator. |
- * @return {@code true} if this type of token represents an additive operator |
- */ |
- bool isAdditiveOperator() => identical(_tokenClass, TokenClass.ADDITIVE_OPERATOR); |
- /** |
- * Return {@code true} if this type of token represents an assignment operator. |
- * @return {@code true} if this type of token represents an assignment operator |
- */ |
- bool isAssignmentOperator() => identical(_tokenClass, TokenClass.ASSIGNMENT_OPERATOR); |
- /** |
- * Return {@code true} if this type of token represents an associative operator. An associative |
- * operator is an operator for which the following equality is true:{@code (a * b) * c == a * (b * c)}. In other words, if the result of applying the operator to |
- * multiple operands does not depend on the order in which those applications occur. |
- * <p> |
- * Note: This method considers the logical-and and logical-or operators to be associative, even |
- * though the order in which the application of those operators can have an effect because |
- * evaluation of the right-hand operand is conditional. |
- * @return {@code true} if this type of token represents an associative operator |
- */ |
- bool isAssociativeOperator() => identical(this, AMPERSAND) || identical(this, AMPERSAND_AMPERSAND) || identical(this, BAR) || identical(this, BAR_BAR) || identical(this, CARET) || identical(this, PLUS) || identical(this, STAR); |
- /** |
- * Return {@code true} if this type of token represents an equality operator. |
- * @return {@code true} if this type of token represents an equality operator |
- */ |
- bool isEqualityOperator() => identical(_tokenClass, TokenClass.EQUALITY_OPERATOR); |
- /** |
- * Return {@code true} if this type of token represents an increment operator. |
- * @return {@code true} if this type of token represents an increment operator |
- */ |
- bool isIncrementOperator() => identical(_lexeme, "++") || identical(_lexeme, "--"); |
- /** |
- * Return {@code true} if this type of token represents a multiplicative operator. |
- * @return {@code true} if this type of token represents a multiplicative operator |
- */ |
- bool isMultiplicativeOperator() => identical(_tokenClass, TokenClass.MULTIPLICATIVE_OPERATOR); |
- /** |
- * Return {@code true} if this token type represents an operator. |
- * @return {@code true} if this token type represents an operator |
- */ |
- bool isOperator() => _tokenClass != TokenClass.NO_CLASS && this != OPEN_PAREN && this != OPEN_SQUARE_BRACKET && this != PERIOD; |
- /** |
- * Return {@code true} if this type of token represents a relational operator. |
- * @return {@code true} if this type of token represents a relational operator |
- */ |
- bool isRelationalOperator() => identical(_tokenClass, TokenClass.RELATIONAL_OPERATOR); |
- /** |
- * Return {@code true} if this type of token represents a shift operator. |
- * @return {@code true} if this type of token represents a shift operator |
- */ |
- bool isShiftOperator() => identical(_tokenClass, TokenClass.SHIFT_OPERATOR); |
- /** |
- * Return {@code true} if this type of token represents a unary postfix operator. |
- * @return {@code true} if this type of token represents a unary postfix operator |
- */ |
- bool isUnaryPostfixOperator() => identical(_tokenClass, TokenClass.UNARY_POSTFIX_OPERATOR); |
- /** |
- * Return {@code true} if this type of token represents a unary prefix operator. |
- * @return {@code true} if this type of token represents a unary prefix operator |
- */ |
- bool isUnaryPrefixOperator() => identical(_tokenClass, TokenClass.UNARY_PREFIX_OPERATOR); |
- /** |
- * Return {@code true} if this token type represents an operator that can be defined by users. |
- * @return {@code true} if this token type represents an operator that can be defined by users |
- */ |
- bool isUserDefinableOperator() => identical(_lexeme, "==") || identical(_lexeme, "~") || identical(_lexeme, "[]") || identical(_lexeme, "[]=") || identical(_lexeme, "*") || identical(_lexeme, "/") || identical(_lexeme, "%") || identical(_lexeme, "~/") || identical(_lexeme, "+") || identical(_lexeme, "-") || identical(_lexeme, "<<") || identical(_lexeme, ">>") || identical(_lexeme, ">=") || identical(_lexeme, ">") || identical(_lexeme, "<=") || identical(_lexeme, "<") || identical(_lexeme, "&") || identical(_lexeme, "^") || identical(_lexeme, "|"); |
- String toString() => __name; |
-} |
-class TokenType_EOF extends TokenType { |
- TokenType_EOF(String ___name, int ___ordinal, TokenClass arg0, String arg1) : super.con2(___name, ___ordinal, arg0, arg1); |
- String toString() => "-eof-"; |
-} |