/*
* Copyright (c) 2010, 2015, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License version 2 only, as
* published by the Free Software Foundation. Oracle designates this
* particular file as subject to the "Classpath" exception as provided
* by Oracle in the LICENSE file that accompanied this code.
*
* This code is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
* version 2 for more details (a copy is included in the LICENSE file that
* accompanied this code).
*
* You should have received a copy of the GNU General Public License version
* 2 along with this work; if not, write to the Free Software Foundation,
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
*
* Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
* or visit www.oracle.com if you need additional information or have any
* questions.
*/
package jdk.nashorn.internal.parser;
import static jdk.nashorn.internal.parser.TokenType.ADD;
import static jdk.nashorn.internal.parser.TokenType.BINARY_NUMBER;
import static jdk.nashorn.internal.parser.TokenType.COMMENT;
import static jdk.nashorn.internal.parser.TokenType.DECIMAL;
import static jdk.nashorn.internal.parser.TokenType.DIRECTIVE_COMMENT;
import static jdk.nashorn.internal.parser.TokenType.EOF;
import static jdk.nashorn.internal.parser.TokenType.EOL;
import static jdk.nashorn.internal.parser.TokenType.ERROR;
import static jdk.nashorn.internal.parser.TokenType.ESCSTRING;
import static jdk.nashorn.internal.parser.TokenType.EXECSTRING;
import static jdk.nashorn.internal.parser.TokenType.FLOATING;
import static jdk.nashorn.internal.parser.TokenType.FUNCTION;
import static jdk.nashorn.internal.parser.TokenType.HEXADECIMAL;
import static jdk.nashorn.internal.parser.TokenType.LBRACE;
import static jdk.nashorn.internal.parser.TokenType.LPAREN;
import static jdk.nashorn.internal.parser.TokenType.OCTAL;
import static jdk.nashorn.internal.parser.TokenType.OCTAL_LEGACY;
import static jdk.nashorn.internal.parser.TokenType.RBRACE;
import static jdk.nashorn.internal.parser.TokenType.REGEX;
import static jdk.nashorn.internal.parser.TokenType.RPAREN;
import static jdk.nashorn.internal.parser.TokenType.STRING;
import static jdk.nashorn.internal.parser.TokenType.TEMPLATE;
import static jdk.nashorn.internal.parser.TokenType.TEMPLATE_HEAD;
import static jdk.nashorn.internal.parser.TokenType.TEMPLATE_MIDDLE;
import static jdk.nashorn.internal.parser.TokenType.TEMPLATE_TAIL;
import static jdk.nashorn.internal.parser.TokenType.XML;
import java.io.Serializable;
import jdk.nashorn.internal.runtime.ECMAErrors;
import jdk.nashorn.internal.runtime.ErrorManager;
import jdk.nashorn.internal.runtime.JSErrorType;
import jdk.nashorn.internal.runtime.JSType;
import jdk.nashorn.internal.runtime.ParserException;
import jdk.nashorn.internal.runtime.Source;
import jdk.nashorn.internal.runtime.options.Options;
Responsible for converting source content into a stream of tokens.
/**
* Responsible for converting source content into a stream of tokens.
*
*/
@SuppressWarnings("fallthrough")
public class Lexer extends Scanner {
private static final long MIN_INT_L = Integer.MIN_VALUE;
private static final long MAX_INT_L = Integer.MAX_VALUE;
private static final boolean XML_LITERALS = Options.getBooleanProperty("nashorn.lexer.xmlliterals");
Content source. /** Content source. */
private final Source source;
Buffered stream for tokens. /** Buffered stream for tokens. */
private final TokenStream stream;
True if here and edit strings are supported. /** True if here and edit strings are supported. */
private final boolean scripting;
True if parsing in ECMAScript 6 mode. /** True if parsing in ECMAScript 6 mode. */
private final boolean es6;
True if a nested scan. (scan to completion, no EOF.) /** True if a nested scan. (scan to completion, no EOF.) */
private final boolean nested;
Pending new line number and position. /** Pending new line number and position. */
int pendingLine;
Position of last EOL + 1. /** Position of last EOL + 1. */
private int linePosition;
Type of last token added. /** Type of last token added. */
private TokenType last;
private final boolean pauseOnFunctionBody;
private boolean pauseOnNextLeftBrace;
private int templateExpressionOpenBraces;
private static final String JAVASCRIPT_OTHER_WHITESPACE =
"\u2028" + // line separator
"\u2029" + // paragraph separator
"\u00a0" + // Latin-1 space
"\u1680" + // Ogham space mark
"\u180e" + // separator, Mongolian vowel
"\u2000" + // en quad
"\u2001" + // em quad
"\u2002" + // en space
"\u2003" + // em space
"\u2004" + // three-per-em space
"\u2005" + // four-per-em space
"\u2006" + // six-per-em space
"\u2007" + // figure space
"\u2008" + // punctuation space
"\u2009" + // thin space
"\u200a" + // hair space
"\u202f" + // narrow no-break space
"\u205f" + // medium mathematical space
"\u3000" + // ideographic space
"\ufeff" // byte order mark
;
private static final String JAVASCRIPT_WHITESPACE_IN_REGEXP =
"\\u000a" + // line feed
"\\u000d" + // carriage return (ctrl-m)
"\\u2028" + // line separator
"\\u2029" + // paragraph separator
"\\u0009" + // tab
"\\u0020" + // ASCII space
"\\u000b" + // tabulation line
"\\u000c" + // ff (ctrl-l)
"\\u00a0" + // Latin-1 space
"\\u1680" + // Ogham space mark
"\\u180e" + // separator, Mongolian vowel
"\\u2000" + // en quad
"\\u2001" + // em quad
"\\u2002" + // en space
"\\u2003" + // em space
"\\u2004" + // three-per-em space
"\\u2005" + // four-per-em space
"\\u2006" + // six-per-em space
"\\u2007" + // figure space
"\\u2008" + // punctuation space
"\\u2009" + // thin space
"\\u200a" + // hair space
"\\u202f" + // narrow no-break space
"\\u205f" + // medium mathematical space
"\\u3000" + // ideographic space
"\\ufeff" // byte order mark
;
static String unicodeEscape(final char ch) {
final StringBuilder sb = new StringBuilder();
sb.append("\\u");
final String hex = Integer.toHexString(ch);
for (int i = hex.length(); i < 4; i++) {
sb.append('0');
}
sb.append(hex);
return sb.toString();
}
Constructor
Params: - source – the source
- stream – the token stream to lex
/**
* Constructor
*
* @param source the source
* @param stream the token stream to lex
*/
public Lexer(final Source source, final TokenStream stream) {
this(source, stream, false, false);
}
Constructor
Params: - source – the source
- stream – the token stream to lex
- scripting – are we in scripting mode
- es6 – are we in ECMAScript 6 mode
/**
* Constructor
*
* @param source the source
* @param stream the token stream to lex
* @param scripting are we in scripting mode
* @param es6 are we in ECMAScript 6 mode
*/
public Lexer(final Source source, final TokenStream stream, final boolean scripting, final boolean es6) {
this(source, 0, source.getLength(), stream, scripting, es6, false);
}
Constructor
Params: - source – the source
- start – start position in source from which to start lexing
- len – length of source segment to lex
- stream – token stream to lex
- scripting – are we in scripting mode
- es6 – are we in ECMAScript 6 mode
- pauseOnFunctionBody – if true, lexer will return from
lexify()
when it encounters a function body. This is used with the feature where the parser is skipping nested function bodies to avoid reading ahead unnecessarily when we skip the function bodies.
/**
* Constructor
*
* @param source the source
* @param start start position in source from which to start lexing
* @param len length of source segment to lex
* @param stream token stream to lex
* @param scripting are we in scripting mode
* @param es6 are we in ECMAScript 6 mode
* @param pauseOnFunctionBody if true, lexer will return from {@link #lexify()} when it encounters a
* function body. This is used with the feature where the parser is skipping nested function bodies to
* avoid reading ahead unnecessarily when we skip the function bodies.
*/
public Lexer(final Source source, final int start, final int len, final TokenStream stream, final boolean scripting, final boolean es6, final boolean pauseOnFunctionBody) {
super(source.getContent(), 1, start, len);
this.source = source;
this.stream = stream;
this.scripting = scripting;
this.es6 = es6;
this.nested = false;
this.pendingLine = 1;
this.last = EOL;
this.pauseOnFunctionBody = pauseOnFunctionBody;
}
private Lexer(final Lexer lexer, final State state) {
super(lexer, state);
source = lexer.source;
stream = lexer.stream;
scripting = lexer.scripting;
es6 = lexer.es6;
nested = true;
pendingLine = state.pendingLine;
linePosition = state.linePosition;
last = EOL;
pauseOnFunctionBody = false;
}
static class State extends Scanner.State {
Pending new line number and position. /** Pending new line number and position. */
public final int pendingLine;
Position of last EOL + 1. /** Position of last EOL + 1. */
public final int linePosition;
Type of last token added. /** Type of last token added. */
public final TokenType last;
/*
* Constructor.
*/
State(final int position, final int limit, final int line, final int pendingLine, final int linePosition, final TokenType last) {
super(position, limit, line);
this.pendingLine = pendingLine;
this.linePosition = linePosition;
this.last = last;
}
}
Save the state of the scan.
Returns: Captured state.
/**
* Save the state of the scan.
*
* @return Captured state.
*/
@Override
State saveState() {
return new State(position, limit, line, pendingLine, linePosition, last);
}
Restore the state of the scan.
Params: - state –
Captured state.
/**
* Restore the state of the scan.
*
* @param state
* Captured state.
*/
void restoreState(final State state) {
super.restoreState(state);
pendingLine = state.pendingLine;
linePosition = state.linePosition;
last = state.last;
}
Add a new token to the stream.
Params: - type –
Token type.
- start –
Start position.
- end –
End position.
/**
* Add a new token to the stream.
*
* @param type
* Token type.
* @param start
* Start position.
* @param end
* End position.
*/
protected void add(final TokenType type, final int start, final int end) {
// Record last token.
last = type;
// Only emit the last EOL in a cluster.
if (type == EOL) {
pendingLine = end;
linePosition = start;
} else {
// Write any pending EOL to stream.
if (pendingLine != -1) {
stream.put(Token.toDesc(EOL, linePosition, pendingLine));
pendingLine = -1;
}
// Write token to stream.
stream.put(Token.toDesc(type, start, end - start));
}
}
Add a new token to the stream.
Params: - type –
Token type.
- start –
Start position.
/**
* Add a new token to the stream.
*
* @param type
* Token type.
* @param start
* Start position.
*/
protected void add(final TokenType type, final int start) {
add(type, start, position);
}
Return the String of valid whitespace characters for regular
expressions in JavaScript
Returns: regexp whitespace string
/**
* Return the String of valid whitespace characters for regular
* expressions in JavaScript
* @return regexp whitespace string
*/
public static String getWhitespaceRegExp() {
return JAVASCRIPT_WHITESPACE_IN_REGEXP;
}
Skip end of line.
Params: - addEOL – true if EOL token should be recorded.
/**
* Skip end of line.
*
* @param addEOL true if EOL token should be recorded.
*/
private void skipEOL(final boolean addEOL) {
if (ch0 == '\r') { // detect \r\n pattern
skip(1);
if (ch0 == '\n') {
skip(1);
}
} else { // all other space, ch0 is guaranteed to be EOL or \0
skip(1);
}
// bump up line count
line++;
if (addEOL) {
// Add an EOL token.
add(EOL, position, line);
}
}
Skip over rest of line including end of line.
Params: - addEOL – true if EOL token should be recorded.
/**
* Skip over rest of line including end of line.
*
* @param addEOL true if EOL token should be recorded.
*/
private void skipLine(final boolean addEOL) {
// Ignore characters.
while (!isEOL(ch0) && !atEOF()) {
skip(1);
}
// Skip over end of line.
skipEOL(addEOL);
}
Test whether a char is valid JavaScript whitespace
Params: - ch – a char
Returns: true if valid JavaScript whitespace
/**
* Test whether a char is valid JavaScript whitespace
* @param ch a char
* @return true if valid JavaScript whitespace
*/
public static boolean isJSWhitespace(final char ch) {
return ch == ' ' // space
|| ch >= '\t' && ch <= '\r' // 0x09..0x0d: tab, line feed, tabulation line, ff, carriage return
|| ch >= 160 && isOtherJSWhitespace(ch);
}
private static boolean isOtherJSWhitespace(final char ch) {
return JAVASCRIPT_OTHER_WHITESPACE.indexOf(ch) != -1;
}
Test whether a char is valid JavaScript end of line
Params: - ch – a char
Returns: true if valid JavaScript end of line
/**
* Test whether a char is valid JavaScript end of line
* @param ch a char
* @return true if valid JavaScript end of line
*/
public static boolean isJSEOL(final char ch) {
return ch == '\n' // line feed
|| ch == '\r' // carriage return (ctrl-m)
|| ch == '\u2028' // line separator
|| ch == '\u2029'; // paragraph separator
}
Test if char is a string delimiter, e.g. '\' or '"'.
Params: - ch – a char
Returns: true if string delimiter
/**
* Test if char is a string delimiter, e.g. '\' or '"'.
* @param ch a char
* @return true if string delimiter
*/
protected boolean isStringDelimiter(final char ch) {
return ch == '\'' || ch == '"';
}
Test if char is a template literal delimiter ('`').
/**
* Test if char is a template literal delimiter ('`').
*/
private static boolean isTemplateDelimiter(final char ch) {
return ch == '`';
}
Test whether a char is valid JavaScript whitespace
Params: - ch – a char
Returns: true if valid JavaScript whitespace
/**
* Test whether a char is valid JavaScript whitespace
* @param ch a char
* @return true if valid JavaScript whitespace
*/
protected boolean isWhitespace(final char ch) {
return Lexer.isJSWhitespace(ch);
}
Test whether a char is valid JavaScript end of line
Params: - ch – a char
Returns: true if valid JavaScript end of line
/**
* Test whether a char is valid JavaScript end of line
* @param ch a char
* @return true if valid JavaScript end of line
*/
protected boolean isEOL(final char ch) {
return Lexer.isJSEOL(ch);
}
Skip over whitespace and detect end of line, adding EOL tokens if
encountered.
Params: - addEOL – true if EOL tokens should be recorded.
/**
* Skip over whitespace and detect end of line, adding EOL tokens if
* encountered.
*
* @param addEOL true if EOL tokens should be recorded.
*/
private void skipWhitespace(final boolean addEOL) {
while (isWhitespace(ch0)) {
if (isEOL(ch0)) {
skipEOL(addEOL);
} else {
skip(1);
}
}
}
Skip over comments.
Returns: True if a comment.
/**
* Skip over comments.
*
* @return True if a comment.
*/
protected boolean skipComments() {
// Save the current position.
final int start = position;
if (ch0 == '/') {
// Is it a // comment.
if (ch1 == '/') {
// Skip over //.
skip(2);
boolean directiveComment = false;
if ((ch0 == '#' || ch0 == '@') && (ch1 == ' ')) {
directiveComment = true;
}
// Scan for EOL.
while (!atEOF() && !isEOL(ch0)) {
skip(1);
}
// Did detect a comment.
add(directiveComment? DIRECTIVE_COMMENT : COMMENT, start);
return true;
} else if (ch1 == '*') {
// Skip over /*.
skip(2);
// Scan for */.
while (!atEOF() && !(ch0 == '*' && ch1 == '/')) {
// If end of line handle else skip character.
if (isEOL(ch0)) {
skipEOL(true);
} else {
skip(1);
}
}
if (atEOF()) {
// TODO - Report closing */ missing in parser.
add(ERROR, start);
} else {
// Skip */.
skip(2);
}
// Did detect a comment.
add(COMMENT, start);
return true;
}
} else if (ch0 == '#') {
assert scripting;
// shell style comment
// Skip over #.
skip(1);
// Scan for EOL.
while (!atEOF() && !isEOL(ch0)) {
skip(1);
}
// Did detect a comment.
add(COMMENT, start);
return true;
}
// Not a comment.
return false;
}
Convert a regex token to a token object.
Params: - start – Position in source content.
- length – Length of regex token.
Returns: Regex token object.
/**
* Convert a regex token to a token object.
*
* @param start Position in source content.
* @param length Length of regex token.
* @return Regex token object.
*/
public RegexToken valueOfPattern(final int start, final int length) {
// Save the current position.
final int savePosition = position;
// Reset to beginning of content.
reset(start);
// Buffer for recording characters.
final StringBuilder sb = new StringBuilder(length);
// Skip /.
skip(1);
boolean inBrackets = false;
// Scan for closing /, stopping at end of line.
while (!atEOF() && ch0 != '/' && !isEOL(ch0) || inBrackets) {
// Skip over escaped character.
if (ch0 == '\\') {
sb.append(ch0);
sb.append(ch1);
skip(2);
} else {
if (ch0 == '[') {
inBrackets = true;
} else if (ch0 == ']') {
inBrackets = false;
}
// Skip literal character.
sb.append(ch0);
skip(1);
}
}
// Get pattern as string.
final String regex = sb.toString();
// Skip /.
skip(1);
// Options as string.
final String options = source.getString(position, scanIdentifier());
reset(savePosition);
// Compile the pattern.
return new RegexToken(regex, options);
}
Return true if the given token can be the beginning of a literal.
Params: - token – a token
Returns: true if token can start a literal.
/**
* Return true if the given token can be the beginning of a literal.
*
* @param token a token
* @return true if token can start a literal.
*/
public boolean canStartLiteral(final TokenType token) {
return token.startsWith('/') || ((scripting || XML_LITERALS) && token.startsWith('<'));
}
interface to receive line information for multi-line literals.
/**
* interface to receive line information for multi-line literals.
*/
protected interface LineInfoReceiver {
Receives line information
Params: - line – last line number
- linePosition – position of last line
/**
* Receives line information
* @param line last line number
* @param linePosition position of last line
*/
public void lineInfo(int line, int linePosition);
}
Check whether the given token represents the beginning of a literal. If so scan
the literal and return true, otherwise return false.
Params: - token – the token.
- startTokenType – the token type.
- lir – LineInfoReceiver that receives line info for multi-line string literals.
Returns: True if a literal beginning with startToken was found and scanned.
/**
* Check whether the given token represents the beginning of a literal. If so scan
* the literal and return <tt>true</tt>, otherwise return false.
*
* @param token the token.
* @param startTokenType the token type.
* @param lir LineInfoReceiver that receives line info for multi-line string literals.
* @return True if a literal beginning with startToken was found and scanned.
*/
protected boolean scanLiteral(final long token, final TokenType startTokenType, final LineInfoReceiver lir) {
// Check if it can be a literal.
if (!canStartLiteral(startTokenType)) {
return false;
}
// We break on ambiguous tokens so if we already moved on it can't be a literal.
if (stream.get(stream.last()) != token) {
return false;
}
// Record current position in case multiple heredocs start on this line - see JDK-8073653
final State state = saveState();
// Rewind to token start position
reset(Token.descPosition(token));
if (ch0 == '/') {
return scanRegEx();
} else if (ch0 == '<') {
if (ch1 == '<') {
return scanHereString(lir, state);
} else if (Character.isJavaIdentifierStart(ch1)) {
return scanXMLLiteral();
}
}
return false;
}
Scan over regex literal.
Returns: True if a regex literal.
/**
* Scan over regex literal.
*
* @return True if a regex literal.
*/
private boolean scanRegEx() {
assert ch0 == '/';
// Make sure it's not a comment.
if (ch1 != '/' && ch1 != '*') {
// Record beginning of literal.
final int start = position;
// Skip /.
skip(1);
boolean inBrackets = false;
// Scan for closing /, stopping at end of line.
while (!atEOF() && (ch0 != '/' || inBrackets) && !isEOL(ch0)) {
// Skip over escaped character.
if (ch0 == '\\') {
skip(1);
if (isEOL(ch0)) {
reset(start);
return false;
}
skip(1);
} else {
if (ch0 == '[') {
inBrackets = true;
} else if (ch0 == ']') {
inBrackets = false;
}
// Skip literal character.
skip(1);
}
}
// If regex literal.
if (ch0 == '/') {
// Skip /.
skip(1);
// Skip over options.
while (!atEOF() && Character.isJavaIdentifierPart(ch0) || ch0 == '\\' && ch1 == 'u') {
skip(1);
}
// Add regex token.
add(REGEX, start);
// Regex literal detected.
return true;
}
// False start try again.
reset(start);
}
// Regex literal not detected.
return false;
}
Convert a digit to a integer. Can't use Character.digit since we are
restricted to ASCII by the spec.
Params: - ch – Character to convert.
- base – Numeric base.
Returns: The converted digit or -1 if invalid.
/**
* Convert a digit to a integer. Can't use Character.digit since we are
* restricted to ASCII by the spec.
*
* @param ch Character to convert.
* @param base Numeric base.
*
* @return The converted digit or -1 if invalid.
*/
protected static int convertDigit(final char ch, final int base) {
int digit;
if ('0' <= ch && ch <= '9') {
digit = ch - '0';
} else if ('A' <= ch && ch <= 'Z') {
digit = ch - 'A' + 10;
} else if ('a' <= ch && ch <= 'z') {
digit = ch - 'a' + 10;
} else {
return -1;
}
return digit < base ? digit : -1;
}
Get the value of a hexadecimal numeric sequence.
Params: - length – Number of digits.
- type – Type of token to report against.
Returns: Value of sequence or < 0 if no digits.
/**
* Get the value of a hexadecimal numeric sequence.
*
* @param length Number of digits.
* @param type Type of token to report against.
* @return Value of sequence or < 0 if no digits.
*/
private int hexSequence(final int length, final TokenType type) {
int value = 0;
for (int i = 0; i < length; i++) {
final int digit = convertDigit(ch0, 16);
if (digit == -1) {
error(Lexer.message("invalid.hex"), type, position, limit);
return i == 0 ? -1 : value;
}
value = digit | value << 4;
skip(1);
}
return value;
}
Get the value of an octal numeric sequence. This parses up to 3 digits with a maximum value of 255.
Returns: Value of sequence.
/**
* Get the value of an octal numeric sequence. This parses up to 3 digits with a maximum value of 255.
*
* @return Value of sequence.
*/
private int octalSequence() {
int value = 0;
for (int i = 0; i < 3; i++) {
final int digit = convertDigit(ch0, 8);
if (digit == -1) {
break;
}
value = digit | value << 3;
skip(1);
if (i == 1 && value >= 32) {
break;
}
}
return value;
}
Convert a string to a JavaScript identifier.
Params: - start – Position in source content.
- length – Length of token.
Returns: Ident string or null if an error.
/**
* Convert a string to a JavaScript identifier.
*
* @param start Position in source content.
* @param length Length of token.
* @return Ident string or null if an error.
*/
private String valueOfIdent(final int start, final int length) throws RuntimeException {
// Save the current position.
final int savePosition = position;
// End of scan.
final int end = start + length;
// Reset to beginning of content.
reset(start);
// Buffer for recording characters.
final StringBuilder sb = new StringBuilder(length);
// Scan until end of line or end of file.
while (!atEOF() && position < end && !isEOL(ch0)) {
// If escape character.
if (ch0 == '\\' && ch1 == 'u') {
skip(2);
final int ch = hexSequence(4, TokenType.IDENT);
if (isWhitespace((char)ch)) {
return null;
}
if (ch < 0) {
sb.append('\\');
sb.append('u');
} else {
sb.append((char)ch);
}
} else {
// Add regular character.
sb.append(ch0);
skip(1);
}
}
// Restore position.
reset(savePosition);
return sb.toString();
}
Scan over and identifier or keyword. Handles identifiers containing
encoded Unicode chars.
Example:
var \u0042 = 44;
/**
* Scan over and identifier or keyword. Handles identifiers containing
* encoded Unicode chars.
*
* Example:
*
* var \u0042 = 44;
*/
private void scanIdentifierOrKeyword() {
// Record beginning of identifier.
final int start = position;
// Scan identifier.
final int length = scanIdentifier();
// Check to see if it is a keyword.
final TokenType type = TokenLookup.lookupKeyword(content, start, length);
if (type == FUNCTION && pauseOnFunctionBody) {
pauseOnNextLeftBrace = true;
}
// Add keyword or identifier token.
add(type, start);
}
Convert a string to a JavaScript string object.
Params: - start – Position in source content.
- length – Length of token.
Returns: JavaScript string object.
/**
* Convert a string to a JavaScript string object.
*
* @param start Position in source content.
* @param length Length of token.
* @return JavaScript string object.
*/
private String valueOfString(final int start, final int length, final boolean strict) throws RuntimeException {
// Save the current position.
final int savePosition = position;
// Calculate the end position.
final int end = start + length;
// Reset to beginning of string.
reset(start);
// Buffer for recording characters.
final StringBuilder sb = new StringBuilder(length);
// Scan until end of string.
while (position < end) {
// If escape character.
if (ch0 == '\\') {
skip(1);
final char next = ch0;
final int afterSlash = position;
skip(1);
// Special characters.
switch (next) {
case '0':
case '1':
case '2':
case '3':
case '4':
case '5':
case '6':
case '7': {
if (strict) {
// "\0" itself is allowed in strict mode. Only other 'real'
// octal escape sequences are not allowed (eg. "\02", "\31").
// See section 7.8.4 String literals production EscapeSequence
if (next != '0' || (ch0 >= '0' && ch0 <= '9')) {
error(Lexer.message("strict.no.octal"), STRING, position, limit);
}
}
reset(afterSlash);
// Octal sequence.
final int ch = octalSequence();
if (ch < 0) {
sb.append('\\');
sb.append('x');
} else {
sb.append((char)ch);
}
break;
}
case 'n':
sb.append('\n');
break;
case 't':
sb.append('\t');
break;
case 'b':
sb.append('\b');
break;
case 'f':
sb.append('\f');
break;
case 'r':
sb.append('\r');
break;
case '\'':
sb.append('\'');
break;
case '\"':
sb.append('\"');
break;
case '\\':
sb.append('\\');
break;
case '\r': // CR | CRLF
if (ch0 == '\n') {
skip(1);
}
// fall through
case '\n': // LF
case '\u2028': // LS
case '\u2029': // PS
// continue on the next line, slash-return continues string
// literal
break;
case 'x': {
// Hex sequence.
final int ch = hexSequence(2, STRING);
if (ch < 0) {
sb.append('\\');
sb.append('x');
} else {
sb.append((char)ch);
}
}
break;
case 'u': {
// Unicode sequence.
final int ch = hexSequence(4, STRING);
if (ch < 0) {
sb.append('\\');
sb.append('u');
} else {
sb.append((char)ch);
}
}
break;
case 'v':
sb.append('\u000B');
break;
// All other characters.
default:
sb.append(next);
break;
}
} else if (ch0 == '\r') {
// Convert CR-LF or CR to LF line terminator.
sb.append('\n');
skip(ch1 == '\n' ? 2 : 1);
} else {
// Add regular character.
sb.append(ch0);
skip(1);
}
}
// Restore position.
reset(savePosition);
return sb.toString();
}
Scan over a string literal.
Params: - add – true if we are not just scanning but should actually modify the token stream
/**
* Scan over a string literal.
* @param add true if we are not just scanning but should actually modify the token stream
*/
protected void scanString(final boolean add) {
// Type of string.
TokenType type = STRING;
// Record starting quote.
final char quote = ch0;
// Skip over quote.
skip(1);
// Record beginning of string content.
final State stringState = saveState();
// Scan until close quote or end of line.
while (!atEOF() && ch0 != quote && !isEOL(ch0)) {
// Skip over escaped character.
if (ch0 == '\\') {
type = ESCSTRING;
skip(1);
if (! isEscapeCharacter(ch0)) {
error(Lexer.message("invalid.escape.char"), STRING, position, limit);
}
if (isEOL(ch0)) {
// Multiline string literal
skipEOL(false);
continue;
}
}
// Skip literal character.
skip(1);
}
// If close quote.
if (ch0 == quote) {
// Skip close quote.
skip(1);
} else {
error(Lexer.message("missing.close.quote"), STRING, position, limit);
}
// If not just scanning.
if (add) {
// Record end of string.
stringState.setLimit(position - 1);
if (scripting && !stringState.isEmpty()) {
switch (quote) {
case '`':
// Mark the beginning of an exec string.
add(EXECSTRING, stringState.position, stringState.limit);
// Frame edit string with left brace.
add(LBRACE, stringState.position, stringState.position);
// Process edit string.
editString(type, stringState);
// Frame edit string with right brace.
add(RBRACE, stringState.limit, stringState.limit);
break;
case '"':
// Only edit double quoted strings.
editString(type, stringState);
break;
case '\'':
// Add string token without editing.
add(type, stringState.position, stringState.limit);
break;
default:
break;
}
} else {
/// Add string token without editing.
add(type, stringState.position, stringState.limit);
}
}
}
Scan over a template string literal.
/**
* Scan over a template string literal.
*/
private void scanTemplate() {
assert ch0 == '`';
TokenType type = TEMPLATE;
// Skip over quote and record beginning of string content.
skip(1);
State stringState = saveState();
// Scan until close quote
while (!atEOF()) {
// Skip over escaped character.
if (ch0 == '`') {
skip(1);
// Record end of string.
stringState.setLimit(position - 1);
add(type == TEMPLATE ? type : TEMPLATE_TAIL, stringState.position, stringState.limit);
return;
} else if (ch0 == '$' && ch1 == '{') {
skip(2);
stringState.setLimit(position - 2);
add(type == TEMPLATE ? TEMPLATE_HEAD : type, stringState.position, stringState.limit);
// scan to RBRACE
final Lexer expressionLexer = new Lexer(this, saveState());
expressionLexer.templateExpressionOpenBraces = 1;
expressionLexer.lexify();
restoreState(expressionLexer.saveState());
// scan next middle or tail of the template literal
assert ch0 == '}';
type = TEMPLATE_MIDDLE;
// Skip over rbrace and record beginning of string content.
skip(1);
stringState = saveState();
continue;
} else if (ch0 == '\\') {
skip(1);
// EscapeSequence
if (!isEscapeCharacter(ch0)) {
error(Lexer.message("invalid.escape.char"), TEMPLATE, position, limit);
}
if (isEOL(ch0)) {
// LineContinuation
skipEOL(false);
continue;
}
} else if (isEOL(ch0)) {
// LineTerminatorSequence
skipEOL(false);
continue;
}
// Skip literal character.
skip(1);
}
error(Lexer.message("missing.close.quote"), TEMPLATE, position, limit);
}
Is the given character a valid escape char after "\" ?
Params: - ch – character to be checked
Returns: if the given character is valid after "\"
/**
* Is the given character a valid escape char after "\" ?
*
* @param ch character to be checked
* @return if the given character is valid after "\"
*/
protected boolean isEscapeCharacter(final char ch) {
return true;
}
Convert string to number.
Params: - valueString – String to convert.
- radix – Numeric base.
Returns: Converted number.
/**
* Convert string to number.
*
* @param valueString String to convert.
* @param radix Numeric base.
* @return Converted number.
*/
private static Number valueOf(final String valueString, final int radix) throws NumberFormatException {
try {
return Integer.parseInt(valueString, radix);
} catch (final NumberFormatException e) {
if (radix == 10) {
return Double.valueOf(valueString);
}
double value = 0.0;
for (int i = 0; i < valueString.length(); i++) {
final char ch = valueString.charAt(i);
// Preverified, should always be a valid digit.
final int digit = convertDigit(ch, radix);
value *= radix;
value += digit;
}
return value;
}
}
Scan a number.
/**
* Scan a number.
*/
protected void scanNumber() {
// Record beginning of number.
final int start = position;
// Assume value is a decimal.
TokenType type = DECIMAL;
// First digit of number.
int digit = convertDigit(ch0, 10);
// If number begins with 0x.
if (digit == 0 && (ch1 == 'x' || ch1 == 'X') && convertDigit(ch2, 16) != -1) {
// Skip over 0xN.
skip(3);
// Skip over remaining digits.
while (convertDigit(ch0, 16) != -1) {
skip(1);
}
type = HEXADECIMAL;
} else if (digit == 0 && es6 && (ch1 == 'o' || ch1 == 'O') && convertDigit(ch2, 8) != -1) {
// Skip over 0oN.
skip(3);
// Skip over remaining digits.
while (convertDigit(ch0, 8) != -1) {
skip(1);
}
type = OCTAL;
} else if (digit == 0 && es6 && (ch1 == 'b' || ch1 == 'B') && convertDigit(ch2, 2) != -1) {
// Skip over 0bN.
skip(3);
// Skip over remaining digits.
while (convertDigit(ch0, 2) != -1) {
skip(1);
}
type = BINARY_NUMBER;
} else {
// Check for possible octal constant.
boolean octal = digit == 0;
// Skip first digit if not leading '.'.
if (digit != -1) {
skip(1);
}
// Skip remaining digits.
while ((digit = convertDigit(ch0, 10)) != -1) {
// Check octal only digits.
octal = octal && digit < 8;
// Skip digit.
skip(1);
}
if (octal && position - start > 1) {
type = OCTAL_LEGACY;
} else if (ch0 == '.' || ch0 == 'E' || ch0 == 'e') {
// Must be a double.
if (ch0 == '.') {
// Skip period.
skip(1);
// Skip mantissa.
while (convertDigit(ch0, 10) != -1) {
skip(1);
}
}
// Detect exponent.
if (ch0 == 'E' || ch0 == 'e') {
// Skip E.
skip(1);
// Detect and skip exponent sign.
if (ch0 == '+' || ch0 == '-') {
skip(1);
}
// Skip exponent.
while (convertDigit(ch0, 10) != -1) {
skip(1);
}
}
type = FLOATING;
}
}
if (Character.isJavaIdentifierStart(ch0)) {
error(Lexer.message("missing.space.after.number"), type, position, 1);
}
// Add number token.
add(type, start);
}
Convert a regex token to a token object.
Params: - start – Position in source content.
- length – Length of regex token.
Returns: Regex token object.
/**
* Convert a regex token to a token object.
*
* @param start Position in source content.
* @param length Length of regex token.
* @return Regex token object.
*/
XMLToken valueOfXML(final int start, final int length) {
return new XMLToken(source.getString(start, length));
}
Scan over a XML token.
Returns: TRUE if is an XML literal.
/**
* Scan over a XML token.
*
* @return TRUE if is an XML literal.
*/
private boolean scanXMLLiteral() {
assert ch0 == '<' && Character.isJavaIdentifierStart(ch1);
if (XML_LITERALS) {
// Record beginning of xml expression.
final int start = position;
int openCount = 0;
do {
if (ch0 == '<') {
if (ch1 == '/' && Character.isJavaIdentifierStart(ch2)) {
skip(3);
openCount--;
} else if (Character.isJavaIdentifierStart(ch1)) {
skip(2);
openCount++;
} else if (ch1 == '?') {
skip(2);
} else if (ch1 == '!' && ch2 == '-' && ch3 == '-') {
skip(4);
} else {
reset(start);
return false;
}
while (!atEOF() && ch0 != '>') {
if (ch0 == '/' && ch1 == '>') {
openCount--;
skip(1);
break;
} else if (ch0 == '\"' || ch0 == '\'') {
scanString(false);
} else {
skip(1);
}
}
if (ch0 != '>') {
reset(start);
return false;
}
skip(1);
} else if (atEOF()) {
reset(start);
return false;
} else {
skip(1);
}
} while (openCount > 0);
add(XML, start);
return true;
}
return false;
}
Scan over identifier characters.
Returns: Length of identifier or zero if none found.
/**
* Scan over identifier characters.
*
* @return Length of identifier or zero if none found.
*/
private int scanIdentifier() {
final int start = position;
// Make sure first character is valid start character.
if (ch0 == '\\' && ch1 == 'u') {
skip(2);
final int ch = hexSequence(4, TokenType.IDENT);
if (!Character.isJavaIdentifierStart(ch)) {
error(Lexer.message("illegal.identifier.character"), TokenType.IDENT, start, position);
}
} else if (!Character.isJavaIdentifierStart(ch0)) {
// Not an identifier.
return 0;
}
// Make sure remaining characters are valid part characters.
while (!atEOF()) {
if (ch0 == '\\' && ch1 == 'u') {
skip(2);
final int ch = hexSequence(4, TokenType.IDENT);
if (!Character.isJavaIdentifierPart(ch)) {
error(Lexer.message("illegal.identifier.character"), TokenType.IDENT, start, position);
}
} else if (Character.isJavaIdentifierPart(ch0)) {
skip(1);
} else {
break;
}
}
// Length of identifier sequence.
return position - start;
}
Compare two identifiers (in content) for equality.
Params: - aStart – Start of first identifier.
- aLength – Length of first identifier.
- bStart – Start of second identifier.
- bLength – Length of second identifier.
Returns: True if equal.
/**
* Compare two identifiers (in content) for equality.
*
* @param aStart Start of first identifier.
* @param aLength Length of first identifier.
* @param bStart Start of second identifier.
* @param bLength Length of second identifier.
* @return True if equal.
*/
private boolean identifierEqual(final int aStart, final int aLength, final int bStart, final int bLength) {
if (aLength == bLength) {
for (int i = 0; i < aLength; i++) {
if (content[aStart + i] != content[bStart + i]) {
return false;
}
}
return true;
}
return false;
}
Detect if a line starts with a marker identifier.
Params: - identStart – Start of identifier.
- identLength – Length of identifier.
Returns: True if detected.
/**
* Detect if a line starts with a marker identifier.
*
* @param identStart Start of identifier.
* @param identLength Length of identifier.
* @return True if detected.
*/
private boolean hasHereMarker(final int identStart, final int identLength) {
// Skip any whitespace.
skipWhitespace(false);
return identifierEqual(identStart, identLength, position, scanIdentifier());
}
Lexer to service edit strings.
/**
* Lexer to service edit strings.
*/
private static class EditStringLexer extends Lexer {
Type of string literals to emit. /** Type of string literals to emit. */
final TokenType stringType;
/*
* Constructor.
*/
EditStringLexer(final Lexer lexer, final TokenType stringType, final State stringState) {
super(lexer, stringState);
this.stringType = stringType;
}
Lexify the contents of the string.
/**
* Lexify the contents of the string.
*/
@Override
public void lexify() {
// Record start of string position.
int stringStart = position;
// Indicate that the priming first string has not been emitted.
boolean primed = false;
while (true) {
// Detect end of content.
if (atEOF()) {
break;
}
// Honour escapes (should be well formed.)
if (ch0 == '\\' && stringType == ESCSTRING) {
skip(2);
continue;
}
// If start of expression.
if (ch0 == '$' && ch1 == '{') {
if (!primed || stringStart != position) {
if (primed) {
add(ADD, stringStart, stringStart + 1);
}
add(stringType, stringStart, position);
primed = true;
}
// Skip ${
skip(2);
// Save expression state.
final State expressionState = saveState();
// Start with one open brace.
int braceCount = 1;
// Scan for the rest of the string.
while (!atEOF()) {
// If closing brace.
if (ch0 == '}') {
// Break only only if matching brace.
if (--braceCount == 0) {
break;
}
} else if (ch0 == '{') {
// Bump up the brace count.
braceCount++;
}
// Skip to next character.
skip(1);
}
// If braces don't match then report an error.
if (braceCount != 0) {
error(Lexer.message("edit.string.missing.brace"), LBRACE, expressionState.position - 1, 1);
}
// Mark end of expression.
expressionState.setLimit(position);
// Skip closing brace.
skip(1);
// Start next string.
stringStart = position;
// Concatenate expression.
add(ADD, expressionState.position, expressionState.position + 1);
add(LPAREN, expressionState.position, expressionState.position + 1);
// Scan expression.
final Lexer lexer = new Lexer(this, expressionState);
lexer.lexify();
// Close out expression parenthesis.
add(RPAREN, position - 1, position);
continue;
}
// Next character in string.
skip(1);
}
// If there is any unemitted string portion.
if (stringStart != limit) {
// Concatenate remaining string.
if (primed) {
add(ADD, stringStart, 1);
}
add(stringType, stringStart, limit);
}
}
}
Edit string for nested expressions.
Params: - stringType – Type of string literals to emit.
- stringState – State of lexer at start of string.
/**
* Edit string for nested expressions.
*
* @param stringType Type of string literals to emit.
* @param stringState State of lexer at start of string.
*/
private void editString(final TokenType stringType, final State stringState) {
// Use special lexer to scan string.
final EditStringLexer lexer = new EditStringLexer(this, stringType, stringState);
lexer.lexify();
// Need to keep lexer informed.
last = stringType;
}
Scan over a here string.
Returns: TRUE if is a here string.
/**
* Scan over a here string.
*
* @return TRUE if is a here string.
*/
private boolean scanHereString(final LineInfoReceiver lir, final State oldState) {
assert ch0 == '<' && ch1 == '<';
if (scripting) {
// Record beginning of here string.
final State saved = saveState();
// << or <<<
final boolean excludeLastEOL = ch2 != '<';
if (excludeLastEOL) {
skip(2);
} else {
skip(3);
}
// Scan identifier. It might be quoted, indicating that no string editing should take place.
final char quoteChar = ch0;
final boolean noStringEditing = quoteChar == '"' || quoteChar == '\'';
if (noStringEditing) {
skip(1);
}
final int identStart = position;
final int identLength = scanIdentifier();
if (noStringEditing) {
if (ch0 != quoteChar) {
error(Lexer.message("here.non.matching.delimiter"), last, position, position);
restoreState(saved);
return false;
}
skip(1);
}
// Check for identifier.
if (identLength == 0) {
// Treat as shift.
restoreState(saved);
return false;
}
// Record rest of line.
final State restState = saveState();
// keep line number updated
int lastLine = line;
skipLine(false);
lastLine++;
int lastLinePosition = position;
restState.setLimit(position);
if (oldState.position > position) {
restoreState(oldState);
skipLine(false);
}
// Record beginning of string.
final State stringState = saveState();
int stringEnd = position;
// Hunt down marker.
while (!atEOF()) {
// Skip any whitespace.
skipWhitespace(false);
if (hasHereMarker(identStart, identLength)) {
break;
}
skipLine(false);
lastLine++;
lastLinePosition = position;
stringEnd = position;
}
// notify last line information
lir.lineInfo(lastLine, lastLinePosition);
// Record end of string.
stringState.setLimit(stringEnd);
// If marker is missing.
if (stringState.isEmpty() || atEOF()) {
error(Lexer.message("here.missing.end.marker", source.getString(identStart, identLength)), last, position, position);
restoreState(saved);
return false;
}
// Remove last end of line if specified.
if (excludeLastEOL) {
// Handles \n.
if (content[stringEnd - 1] == '\n') {
stringEnd--;
}
// Handles \r and \r\n.
if (content[stringEnd - 1] == '\r') {
stringEnd--;
}
// Update end of string.
stringState.setLimit(stringEnd);
}
// Edit string if appropriate.
if (!noStringEditing && !stringState.isEmpty()) {
editString(STRING, stringState);
} else {
// Add here string.
add(STRING, stringState.position, stringState.limit);
}
// Scan rest of original line.
final Lexer restLexer = new Lexer(this, restState);
restLexer.lexify();
return true;
}
return false;
}
Breaks source content down into lex units, adding tokens to the token
stream. The routine scans until the stream buffer is full. Can be called
repeatedly until EOF is detected.
/**
* Breaks source content down into lex units, adding tokens to the token
* stream. The routine scans until the stream buffer is full. Can be called
* repeatedly until EOF is detected.
*/
public void lexify() {
while (!stream.isFull() || nested) {
// Skip over whitespace.
skipWhitespace(true);
// Detect end of file.
if (atEOF()) {
if (!nested) {
// Add an EOF token at the end.
add(EOF, position);
}
break;
}
// Check for comments. Note that we don't scan for regexp and other literals here as
// we may not have enough context to distinguish them from similar looking operators.
// Instead we break on ambiguous operators below and let the parser decide.
if (ch0 == '/' && skipComments()) {
continue;
}
if (scripting && ch0 == '#' && skipComments()) {
continue;
}
// TokenType for lookup of delimiter or operator.
TokenType type;
if (ch0 == '.' && convertDigit(ch1, 10) != -1) {
// '.' followed by digit.
// Scan and add a number.
scanNumber();
} else if ((type = TokenLookup.lookupOperator(ch0, ch1, ch2, ch3)) != null) {
if (templateExpressionOpenBraces > 0) {
if (type == LBRACE) {
templateExpressionOpenBraces++;
} else if (type == RBRACE) {
if (--templateExpressionOpenBraces == 0) {
break;
}
}
}
// Get the number of characters in the token.
final int typeLength = type.getLength();
// Skip that many characters.
skip(typeLength);
// Add operator token.
add(type, position - typeLength);
// Some operator tokens also mark the beginning of regexp, XML, or here string literals.
// We break to let the parser decide what it is.
if (canStartLiteral(type)) {
break;
} else if (type == LBRACE && pauseOnNextLeftBrace) {
pauseOnNextLeftBrace = false;
break;
}
} else if (Character.isJavaIdentifierStart(ch0) || ch0 == '\\' && ch1 == 'u') {
// Scan and add identifier or keyword.
scanIdentifierOrKeyword();
} else if (isStringDelimiter(ch0)) {
// Scan and add a string.
scanString(true);
} else if (Character.isDigit(ch0)) {
// Scan and add a number.
scanNumber();
} else if (isTemplateDelimiter(ch0) && es6) {
// Scan and add template in ES6 mode.
scanTemplate();
} else if (isTemplateDelimiter(ch0) && scripting) {
// Scan and add an exec string ('`') in scripting mode.
scanString(true);
} else {
// Don't recognize this character.
skip(1);
add(ERROR, position - 1);
}
}
}
Return value of token given its token descriptor.
Params: - token – Token descriptor.
Returns: JavaScript value.
/**
* Return value of token given its token descriptor.
*
* @param token Token descriptor.
* @return JavaScript value.
*/
Object getValueOf(final long token, final boolean strict) {
final int start = Token.descPosition(token);
final int len = Token.descLength(token);
switch (Token.descType(token)) {
case DECIMAL:
return Lexer.valueOf(source.getString(start, len), 10); // number
case HEXADECIMAL:
return Lexer.valueOf(source.getString(start + 2, len - 2), 16); // number
case OCTAL_LEGACY:
return Lexer.valueOf(source.getString(start, len), 8); // number
case OCTAL:
return Lexer.valueOf(source.getString(start + 2, len - 2), 8); // number
case BINARY_NUMBER:
return Lexer.valueOf(source.getString(start + 2, len - 2), 2); // number
case FLOATING:
final String str = source.getString(start, len);
final double value = Double.valueOf(str);
if (str.indexOf('.') != -1) {
return value; //number
}
//anything without an explicit decimal point is still subject to a
//"representable as int or long" check. Then the programmer does not
//explicitly code something as a double. For example new Color(int, int, int)
//and new Color(float, float, float) will get ambiguous for cases like
//new Color(1.0, 1.5, 1.5) if we don't respect the decimal point.
//yet we don't want e.g. 1e6 to be a double unnecessarily
if (JSType.isStrictlyRepresentableAsInt(value)) {
return (int)value;
}
return value;
case STRING:
return source.getString(start, len); // String
case ESCSTRING:
return valueOfString(start, len, strict); // String
case IDENT:
return valueOfIdent(start, len); // String
case REGEX:
return valueOfPattern(start, len); // RegexToken::LexerToken
case TEMPLATE:
case TEMPLATE_HEAD:
case TEMPLATE_MIDDLE:
case TEMPLATE_TAIL:
return valueOfString(start, len, true); // String
case XML:
return valueOfXML(start, len); // XMLToken::LexerToken
case DIRECTIVE_COMMENT:
return source.getString(start, len);
default:
break;
}
return null;
}
Get the raw string value of a template literal string part.
Params: - token – template string token
Returns: raw string
/**
* Get the raw string value of a template literal string part.
*
* @param token template string token
* @return raw string
*/
public String valueOfRawString(final long token) {
final int start = Token.descPosition(token);
final int length = Token.descLength(token);
// Save the current position.
final int savePosition = position;
// Calculate the end position.
final int end = start + length;
// Reset to beginning of string.
reset(start);
// Buffer for recording characters.
final StringBuilder sb = new StringBuilder(length);
// Scan until end of string.
while (position < end) {
if (ch0 == '\r') {
// Convert CR-LF or CR to LF line terminator.
sb.append('\n');
skip(ch1 == '\n' ? 2 : 1);
} else {
// Add regular character.
sb.append(ch0);
skip(1);
}
}
// Restore position.
reset(savePosition);
return sb.toString();
}
Get the correctly localized error message for a given message id format arguments
Params: - msgId – message id
- args – format arguments
Returns: message
/**
* Get the correctly localized error message for a given message id format arguments
* @param msgId message id
* @param args format arguments
* @return message
*/
protected static String message(final String msgId, final String... args) {
return ECMAErrors.getMessage("lexer.error." + msgId, args);
}
Generate a runtime exception
Params: - message – error message
- type – token type
- start – start position of lexed error
- length – length of lexed error
Throws: - ParserException – unconditionally
/**
* Generate a runtime exception
*
* @param message error message
* @param type token type
* @param start start position of lexed error
* @param length length of lexed error
* @throws ParserException unconditionally
*/
protected void error(final String message, final TokenType type, final int start, final int length) throws ParserException {
final long token = Token.toDesc(type, start, length);
final int pos = Token.descPosition(token);
final int lineNum = source.getLine(pos);
final int columnNum = source.getColumn(pos);
final String formatted = ErrorManager.format(message, source, lineNum, columnNum, token);
throw new ParserException(JSErrorType.SYNTAX_ERROR, formatted, source, lineNum, columnNum, token);
}
Helper class for Lexer tokens, e.g XML or RegExp tokens.
This is the abstract superclass
/**
* Helper class for Lexer tokens, e.g XML or RegExp tokens.
* This is the abstract superclass
*/
public static abstract class LexerToken implements Serializable {
private static final long serialVersionUID = 1L;
private final String expression;
Constructor
Params: - expression – token expression
/**
* Constructor
* @param expression token expression
*/
protected LexerToken(final String expression) {
this.expression = expression;
}
Get the expression
Returns: expression
/**
* Get the expression
* @return expression
*/
public String getExpression() {
return expression;
}
}
Temporary container for regular expressions.
/**
* Temporary container for regular expressions.
*/
public static class RegexToken extends LexerToken {
private static final long serialVersionUID = 1L;
Options. /** Options. */
private final String options;
Constructor.
Params: - expression – regexp expression
- options – regexp options
/**
* Constructor.
*
* @param expression regexp expression
* @param options regexp options
*/
public RegexToken(final String expression, final String options) {
super(expression);
this.options = options;
}
Get regexp options
Returns: options
/**
* Get regexp options
* @return options
*/
public String getOptions() {
return options;
}
@Override
public String toString() {
return '/' + getExpression() + '/' + options;
}
}
Temporary container for XML expression.
/**
* Temporary container for XML expression.
*/
public static class XMLToken extends LexerToken {
private static final long serialVersionUID = 1L;
Constructor.
Params: - expression – XML expression
/**
* Constructor.
*
* @param expression XML expression
*/
public XMLToken(final String expression) {
super(expression);
}
}
}