com.google.javascript.jscomp.parsing.parser.Scanner Maven / Gradle / Ivy
Show all versions of com.liferay.frontend.js.minifier
/*
* Copyright 2011 The Closure Compiler Authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package com.google.javascript.jscomp.parsing.parser;
import com.google.errorprone.annotations.FormatMethod;
import com.google.errorprone.annotations.FormatString;
import com.google.javascript.jscomp.parsing.parser.trees.Comment;
import com.google.javascript.jscomp.parsing.parser.util.ErrorReporter;
import com.google.javascript.jscomp.parsing.parser.util.SourcePosition;
import com.google.javascript.jscomp.parsing.parser.util.SourceRange;
import java.util.ArrayList;
/**
* Scans javascript source code into tokens. All entrypoints assume the
* caller is not expecting a regular expression literal except for
* nextRegularExpressionLiteralToken.
*
* 7 Lexical Conventions
*/
public class Scanner {
private final ErrorReporter errorReporter;
private final SourceFile source;
private final ArrayList currentTokens = new ArrayList<>();
private int index;
private final CommentRecorder commentRecorder;
private int typeParameterLevel;
public Scanner(ErrorReporter errorReporter, CommentRecorder commentRecorder,
SourceFile source) {
this(errorReporter, commentRecorder, source, 0);
}
public Scanner(ErrorReporter errorReporter, CommentRecorder commentRecorder,
SourceFile file, int offset) {
this.errorReporter = errorReporter;
this.commentRecorder = commentRecorder;
this.source = file;
this.index = offset;
this.typeParameterLevel = 0;
}
public interface CommentRecorder {
void recordComment(Comment.Type type, SourceRange range, String value);
}
private LineNumberTable getLineNumberTable() {
return this.getFile().lineNumberTable;
}
public SourceFile getFile() {
return source;
}
public int getOffset() {
return currentTokens.isEmpty()
? index
: peekToken().location.start.offset;
}
public void setOffset(int index) {
currentTokens.clear();
this.index = index;
}
public SourcePosition getPosition() {
return getPosition(getOffset());
}
private SourcePosition getPosition(int offset) {
return getLineNumberTable().getSourcePosition(offset);
}
private SourceRange getTokenRange(int startOffset) {
return getLineNumberTable().getSourceRange(startOffset, index);
}
public Token nextToken() {
peekToken();
return currentTokens.remove(0);
}
private void clearTokenLookahead() {
index = getOffset();
currentTokens.clear();
}
public LiteralToken nextRegularExpressionLiteralToken() {
clearTokenLookahead();
int beginToken = index;
// leading '/'
nextChar();
// body
if (!skipRegularExpressionBody()) {
return new LiteralToken(
TokenType.REGULAR_EXPRESSION,
getTokenString(beginToken),
getTokenRange(beginToken));
}
// separating '/'
if (peekChar() != '/') {
reportError("Expected '/' in regular expression literal");
return new LiteralToken(
TokenType.REGULAR_EXPRESSION,
getTokenString(beginToken),
getTokenRange(beginToken));
}
nextChar();
// flags
while (isIdentifierPart(peekChar())) {
nextChar();
}
return new LiteralToken(
TokenType.REGULAR_EXPRESSION,
getTokenString(beginToken),
getTokenRange(beginToken));
}
public LiteralToken nextTemplateLiteralToken() {
Token token = nextToken();
if (isAtEnd() || token.type != TokenType.CLOSE_CURLY) {
reportError(getPosition(index), "Expected '}' after expression in template literal");
}
return nextTemplateLiteralTokenShared(TokenType.TEMPLATE_TAIL, TokenType.TEMPLATE_MIDDLE);
}
private boolean skipRegularExpressionBody() {
if (!isRegularExpressionFirstChar(peekChar())) {
reportError("Expected regular expression first char");
return false;
}
if (!skipRegularExpressionChar()) {
return false;
}
while (!isAtEnd() && isRegularExpressionChar(peekChar())) {
if (!skipRegularExpressionChar()) {
return false;
}
}
return true;
}
private boolean skipRegularExpressionChar() {
switch (peekChar()) {
case '\\':
return skipRegularExpressionBackslashSequence();
case '[':
return skipRegularExpressionClass();
default:
nextChar();
return true;
}
}
private boolean skipRegularExpressionBackslashSequence() {
// TODO(tbreisacher): Warn if this is an unnecessary escape, like we do for string literals.
nextChar();
if (isLineTerminator(peekChar())) {
reportError("New line not allowed in regular expression literal");
return false;
}
nextChar();
return true;
}
private boolean skipRegularExpressionClass() {
nextChar();
while (!isAtEnd() && peekRegularExpressionClassChar()) {
if (!skipRegularExpressionClassChar()) {
return false;
}
}
if (peekChar() != ']') {
reportError("']' expected");
return false;
}
nextChar();
return true;
}
private boolean peekRegularExpressionClassChar() {
return peekChar() != ']' && !isLineTerminator(peekChar());
}
private boolean skipRegularExpressionClassChar() {
if (peek('\\')) {
return skipRegularExpressionBackslashSequence();
}
nextChar();
return true;
}
private static boolean isRegularExpressionFirstChar(char ch) {
return isRegularExpressionChar(ch) && ch != '*';
}
private static boolean isRegularExpressionChar(char ch) {
switch (ch) {
case '/':
return false;
case '\\':
case '[':
return true;
default:
return !isLineTerminator(ch);
}
}
public Token peekToken() {
return peekToken(0);
}
public Token peekToken(int index) {
while (currentTokens.size() <= index) {
currentTokens.add(scanToken());
}
return currentTokens.get(index);
}
private boolean isAtEnd() {
return !isValidIndex(index);
}
private boolean isValidIndex(int index) {
return index >= 0 && index < source.contents.length();
}
// 7.2 White Space
/**
* Returns true if the whitespace that was skipped included any
* line terminators.
*/
private boolean skipWhitespace() {
boolean foundLineTerminator = false;
while (!isAtEnd() && peekWhitespace()) {
if (isLineTerminator(nextChar())) {
foundLineTerminator = true;
}
}
return foundLineTerminator;
}
private boolean peekWhitespace() {
return isWhitespace(peekChar());
}
private static boolean isWhitespace(char ch) {
switch (ch) {
case '\u0009': // Tab
case '\u000B': // Vertical Tab
case '\u000C': // Form Feed
case '\u0020': // Space
case '\u00A0': // No-break space
case '\uFEFF': // Byte Order Mark
case '\n': // Line Feed
case '\r': // Carriage Return
case '\u2028': // Line Separator
case '\u2029': // Paragraph Separator
// TODO: there are other Unicode Category 'Zs' chars that should go here.
return true;
default:
return false;
}
}
// 7.3 Line Terminators
private static boolean isLineTerminator(char ch) {
switch (ch) {
case '\n': // Line Feed
case '\r': // Carriage Return
case '\u2028': // Line Separator
case '\u2029': // Paragraph Separator
return true;
default:
return false;
}
}
// 7.4 Comments
private void skipComments() {
while (skipComment())
{}
}
private boolean skipComment() {
boolean isStartOfLine = skipWhitespace();
if (!isAtEnd()) {
switch (peekChar(0)) {
case '/':
switch (peekChar(1)) {
case '/':
skipSingleLineComment();
return true;
case '*':
skipMultiLineComment();
return true;
default: // fall out
}
break;
case '<':
// Check if this is the start of an HTML comment ("").
// Note that the spec does not require us to check for this case,
// but there is some legacy code that depends on this behavior.
if (isStartOfLine && peekChar(1) == '-' && peekChar(2) == '>') {
reportHtmlCommentWarning();
skipSingleLineComment();
return true;
}
break;
case '#':
if (index == 0 && peekChar(1) == '!') {
skipSingleLineComment(Comment.Type.SHEBANG);
return true;
}
break;
default: // fall out
}
}
return false;
}
private void reportHtmlCommentWarning() {
reportWarning(
"In some cases, '' are treated as a '//' "
+ "for legacy reasons. Removing this from your code is "
+ "safe for all browsers currently in use.");
}
private void skipSingleLineComment() {
skipSingleLineComment(Comment.Type.LINE);
}
private void skipSingleLineComment(Comment.Type type) {
int startOffset = index;
while (!isAtEnd() && !isLineTerminator(peekChar())) {
nextChar();
}
SourceRange range = getLineNumberTable().getSourceRange(startOffset, index);
String value = this.source.contents.substring(startOffset, index);
recordComment(type, range, value);
}
private void recordComment(
Comment.Type type, SourceRange range, String value) {
commentRecorder.recordComment(type, range, value);
}
private void skipMultiLineComment() {
int startOffset = index;
nextChar(); // '/'
nextChar(); // '*'
while (!isAtEnd() && (peekChar() != '*' || peekChar(1) != '/')) {
nextChar();
}
if (!isAtEnd()) {
nextChar();
nextChar();
Comment.Type type = Comment.Type.BLOCK;
if (index - startOffset > 4) {
if (this.source.contents.charAt(startOffset + 2) == '*') {
type = Comment.Type.JSDOC;
} else if (this.source.contents.charAt(startOffset + 2) == '!') {
type = Comment.Type.IMPORTANT;
}
}
SourceRange range = getLineNumberTable().getSourceRange(
startOffset, index);
String value = this.source.contents.substring(
startOffset, index);
recordComment(type, range, value);
} else {
reportError("unterminated comment");
}
}
private Token scanToken() {
skipComments();
int beginToken = index;
if (isAtEnd()) {
return createToken(TokenType.END_OF_FILE, beginToken);
}
char ch = nextChar();
switch (ch) {
case '{': return createToken(TokenType.OPEN_CURLY, beginToken);
case '}': return createToken(TokenType.CLOSE_CURLY, beginToken);
case '(': return createToken(TokenType.OPEN_PAREN, beginToken);
case ')': return createToken(TokenType.CLOSE_PAREN, beginToken);
case '[': return createToken(TokenType.OPEN_SQUARE, beginToken);
case ']': return createToken(TokenType.CLOSE_SQUARE, beginToken);
case '.':
if (isDecimalDigit(peekChar())) {
return scanNumberPostPeriod(beginToken);
}
// Harmony spread operator
if (peek('.') && peekChar(1) == '.') {
nextChar();
nextChar();
return createToken(TokenType.SPREAD, beginToken);
}
return createToken(TokenType.PERIOD, beginToken);
case ';': return createToken(TokenType.SEMI_COLON, beginToken);
case ',': return createToken(TokenType.COMMA, beginToken);
case '~': return createToken(TokenType.TILDE, beginToken);
case '?': return createToken(TokenType.QUESTION, beginToken);
case ':': return createToken(TokenType.COLON, beginToken);
case '<':
switch (peekChar()) {
case '<':
nextChar();
if (peek('=')) {
nextChar();
return createToken(TokenType.LEFT_SHIFT_EQUAL, beginToken);
}
return createToken(TokenType.LEFT_SHIFT, beginToken);
case '=':
nextChar();
return createToken(TokenType.LESS_EQUAL, beginToken);
default:
return createToken(TokenType.OPEN_ANGLE, beginToken);
}
case '>':
if (typeParameterLevel > 0) {
return createToken(TokenType.CLOSE_ANGLE, beginToken);
}
switch (peekChar()) {
case '>':
nextChar();
switch (peekChar()) {
case '=':
nextChar();
return createToken(TokenType.RIGHT_SHIFT_EQUAL, beginToken);
case '>':
nextChar();
if (peek('=')) {
nextChar();
return createToken(TokenType.UNSIGNED_RIGHT_SHIFT_EQUAL, beginToken);
}
return createToken(TokenType.UNSIGNED_RIGHT_SHIFT, beginToken);
default:
return createToken(TokenType.RIGHT_SHIFT, beginToken);
}
case '=':
nextChar();
return createToken(TokenType.GREATER_EQUAL, beginToken);
default:
return createToken(TokenType.CLOSE_ANGLE, beginToken);
}
case '=':
switch (peekChar()) {
case '=':
nextChar();
if (peek('=')) {
nextChar();
return createToken(TokenType.EQUAL_EQUAL_EQUAL, beginToken);
}
return createToken(TokenType.EQUAL_EQUAL, beginToken);
case '>':
nextChar();
return createToken(TokenType.ARROW, beginToken);
default:
return createToken(TokenType.EQUAL, beginToken);
}
case '!':
if (peek('=')) {
nextChar();
if (peek('=')) {
nextChar();
return createToken(TokenType.NOT_EQUAL_EQUAL, beginToken);
}
return createToken(TokenType.NOT_EQUAL, beginToken);
}
return createToken(TokenType.BANG, beginToken);
case '*':
if (peek('=')) {
nextChar();
return createToken(TokenType.STAR_EQUAL, beginToken);
} else if (peek('*')) {
nextChar();
// '**' seen so far
if (peek('=')) {
nextChar();
return createToken(TokenType.STAR_STAR_EQUAL, beginToken);
} else {
return createToken(TokenType.STAR_STAR, beginToken);
}
}
return createToken(TokenType.STAR, beginToken);
case '%':
if (peek('=')) {
nextChar();
return createToken(TokenType.PERCENT_EQUAL, beginToken);
}
return createToken(TokenType.PERCENT, beginToken);
case '^':
if (peek('=')) {
nextChar();
return createToken(TokenType.CARET_EQUAL, beginToken);
}
return createToken(TokenType.CARET, beginToken);
case '/':
if (peek('=')) {
nextChar();
return createToken(TokenType.SLASH_EQUAL, beginToken);
}
return createToken(TokenType.SLASH, beginToken);
case '+':
switch (peekChar()) {
case '+':
nextChar();
return createToken(TokenType.PLUS_PLUS, beginToken);
case '=':
nextChar();
return createToken(TokenType.PLUS_EQUAL, beginToken);
default:
return createToken(TokenType.PLUS, beginToken);
}
case '-':
switch (peekChar()) {
case '-':
nextChar();
return createToken(TokenType.MINUS_MINUS, beginToken);
case '=':
nextChar();
return createToken(TokenType.MINUS_EQUAL, beginToken);
default:
return createToken(TokenType.MINUS, beginToken);
}
case '&':
switch (peekChar()) {
case '&':
nextChar();
return createToken(TokenType.AND, beginToken);
case '=':
nextChar();
return createToken(TokenType.AMPERSAND_EQUAL, beginToken);
default:
return createToken(TokenType.AMPERSAND, beginToken);
}
case '|':
switch (peekChar()) {
case '|':
nextChar();
return createToken(TokenType.OR, beginToken);
case '=':
nextChar();
return createToken(TokenType.BAR_EQUAL, beginToken);
default:
return createToken(TokenType.BAR, beginToken);
}
case '#':
return createToken(TokenType.POUND, beginToken);
// TODO: add NumberToken
// TODO: character following NumericLiteral must not be an IdentifierStart or DecimalDigit
case '0':
return scanPostZero(beginToken);
case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9':
return scanPostDigit(beginToken);
case '"':
case '\'':
return scanStringLiteral(beginToken, ch);
case '`':
return scanTemplateLiteral(beginToken);
default:
return scanIdentifierOrKeyword(beginToken, ch);
}
}
private Token scanNumberPostPeriod(int beginToken) {
skipDecimalDigits();
return scanExponentOfNumericLiteral(beginToken);
}
private Token scanPostDigit(int beginToken) {
skipDecimalDigits();
return scanFractionalNumericLiteral(beginToken);
}
private Token scanPostZero(int beginToken) {
switch (peekChar()) {
case 'b':
case 'B':
// binary
nextChar();
if (!isBinaryDigit(peekChar())) {
reportError("Binary Integer Literal must contain at least one digit");
}
skipBinaryDigits();
return new LiteralToken(
TokenType.NUMBER, getTokenString(beginToken), getTokenRange(beginToken));
case 'o':
case 'O':
// octal
nextChar();
if (!isOctalDigit(peekChar())) {
reportError("Octal Integer Literal must contain at least one digit");
}
skipOctalDigits();
if (peek('8') || peek('9')) {
reportError("Invalid octal digit in octal literal.");
}
return new LiteralToken(
TokenType.NUMBER, getTokenString(beginToken), getTokenRange(beginToken));
case 'x':
case 'X':
nextChar();
if (!peekHexDigit()) {
reportError("Hex Integer Literal must contain at least one digit");
}
skipHexDigits();
return new LiteralToken(
TokenType.NUMBER, getTokenString(beginToken), getTokenRange(beginToken));
case 'e':
case 'E':
return scanExponentOfNumericLiteral(beginToken);
case '.':
return scanFractionalNumericLiteral(beginToken);
case '0': case '1': case '2': case '3': case '4':
case '5': case '6': case '7': case '8': case '9':
skipDecimalDigits();
if (peek('.')) {
nextChar();
skipDecimalDigits();
}
return new LiteralToken(
TokenType.NUMBER, getTokenString(beginToken), getTokenRange(beginToken));
default:
return new LiteralToken(
TokenType.NUMBER, getTokenString(beginToken), getTokenRange(beginToken));
}
}
private Token createToken(TokenType type, int beginToken) {
return new Token(type, getTokenRange(beginToken));
}
private Token scanIdentifierOrKeyword(int beginToken, char ch) {
StringBuilder valueBuilder = new StringBuilder();
valueBuilder.append(ch);
boolean containsUnicodeEscape = ch == '\\';
boolean bracedUnicodeEscape = false;
int unicodeEscapeLen = containsUnicodeEscape ? 1 : 0;
ch = peekChar();
while (isIdentifierPart(ch)
|| ch == '\\'
|| (ch == '{' && unicodeEscapeLen == 2)
|| (ch == '}' && bracedUnicodeEscape)) {
if (ch == '\\') {
containsUnicodeEscape = true;
}
// Update length of current Unicode escape.
if (ch == '\\' || unicodeEscapeLen > 0) {
unicodeEscapeLen++;
}
// Enter Unicode point escape.
if (ch == '{') {
bracedUnicodeEscape = true;
}
// Exit Unicode escape
if (ch == '}' || (unicodeEscapeLen >= 6 && !bracedUnicodeEscape)) {
bracedUnicodeEscape = false;
unicodeEscapeLen = 0;
}
// Add character to token
valueBuilder.append(nextChar());
ch = peekChar();
}
String value = valueBuilder.toString();
// Process unicode escapes.
if (containsUnicodeEscape) {
value = processUnicodeEscapes(value);
if (value == null) {
reportError(
getPosition(index),
"Invalid escape sequence");
return createToken(TokenType.ERROR, beginToken);
}
}
// Check to make sure the first character (or the unicode escape at the
// beginning of the identifier) is a valid identifier start character.
char start = value.charAt(0);
if (!isIdentifierStart(start)) {
reportError(
getPosition(beginToken),
"Character '%c' (U+%04X) is not a valid identifier start char",
start, (int) start);
return createToken(TokenType.ERROR, beginToken);
}
Keywords k = Keywords.get(value);
if (k != null) {
return new Token(k.type, getTokenRange(beginToken));
}
// Intern the value to avoid creating lots of copies of the same string.
return new IdentifierToken(getTokenRange(beginToken), value.intern());
}
/**
* Converts unicode escapes in the given string to the equivalent unicode character. If there are
* no escapes, returns the input unchanged. If there is an invalid escape sequence, returns null.
*/
private static String processUnicodeEscapes(String value) {
while (value.contains("\\")) {
int escapeStart = value.indexOf('\\');
try {
if (value.charAt(escapeStart + 1) != 'u') {
return null;
}
String hexDigits;
int escapeEnd;
if (value.charAt(escapeStart + 2) != '{') {
// Simple escape with exactly four hex digits: \\uXXXX
escapeEnd = escapeStart + 6;
hexDigits = value.substring(escapeStart + 2, escapeEnd);
} else {
// Escape with braces can have any number of hex digits: \\u{XXXXXXX}
escapeEnd = escapeStart + 3;
while (Character.digit(value.charAt(escapeEnd), 0x10) >= 0) {
escapeEnd++;
}
if (value.charAt(escapeEnd) != '}') {
return null;
}
hexDigits = value.substring(escapeStart + 3, escapeEnd);
escapeEnd++;
}
// TODO(mattloring): Allow code points greater than the size of a char
char ch = (char) Integer.parseInt(hexDigits, 0x10);
if (!isIdentifierPart(ch)) {
return null;
}
value = value.substring(0, escapeStart) + ch + value.substring(escapeEnd);
} catch (NumberFormatException | StringIndexOutOfBoundsException e) {
return null;
}
}
return value;
}
private static boolean isIdentifierStart(char ch) {
switch (ch) {
case '$':
case '_':
return true;
default:
// Workaround b/36459436
// When running under GWT, Character.isLetter only handles ASCII
// Angular relies heavily on U+0275 (Latin Barred O)
return ch == 0x0275
// TODO: UnicodeLetter also includes Letter Number (NI)
|| Character.isLetter(ch);
}
}
private static boolean isIdentifierPart(char ch) {
// TODO: identifier part character classes
// CombiningMark
// Non-Spacing mark (Mn)
// Combining spacing mark(Mc)
// Connector punctuation (Pc)
// Zero Width Non-Joiner
// Zero Width Joiner
return isIdentifierStart(ch) || Character.isDigit(ch);
}
private Token scanStringLiteral(int beginIndex, char terminator) {
while (peekStringLiteralChar(terminator)) {
if (!skipStringLiteralChar()) {
return new LiteralToken(
TokenType.STRING, getTokenString(beginIndex), getTokenRange(beginIndex));
}
}
if (peekChar() != terminator) {
reportError(getPosition(beginIndex), "Unterminated string literal");
} else {
nextChar();
}
return new LiteralToken(
TokenType.STRING, getTokenString(beginIndex), getTokenRange(beginIndex));
}
private Token scanTemplateLiteral(int beginIndex) {
if (isAtEnd()) {
reportError(getPosition(beginIndex), "Unterminated template literal");
}
return nextTemplateLiteralTokenShared(
TokenType.NO_SUBSTITUTION_TEMPLATE, TokenType.TEMPLATE_HEAD);
}
private LiteralToken nextTemplateLiteralTokenShared(TokenType endType,
TokenType middleType) {
int beginIndex = index;
skipTemplateCharacters();
if (isAtEnd()) {
reportError(getPosition(beginIndex), "Unterminated template literal");
}
String value = getTokenString(beginIndex);
switch (peekChar()) {
case '`':
nextChar();
return new LiteralToken(endType, value, getTokenRange(beginIndex - 1));
case '$':
nextChar(); // $
nextChar(); // {
return new LiteralToken(middleType, value, getTokenRange(beginIndex - 1));
default: // Should have reported error already
return new LiteralToken(endType, value, getTokenRange(beginIndex - 1));
}
}
private String getTokenString(int beginIndex) {
return this.source.contents.substring(beginIndex, index);
}
private boolean peekStringLiteralChar(char terminator) {
return !isAtEnd() && peekChar() != terminator && !isLineTerminator(peekChar());
}
private boolean skipStringLiteralChar() {
if (peek('\\')) {
return skipStringLiteralEscapeSequence(false);
}
nextChar();
return true;
}
private void skipTemplateCharacters() {
while (!isAtEnd()) {
switch (peekChar()) {
case '`':
return;
case '\\':
skipStringLiteralEscapeSequence(true);
break;
case '$':
if (peekChar(1) == '{') {
return;
}
// Fall through.
default:
nextChar();
}
}
}
@SuppressWarnings("IdentityBinaryExpression") // for "skipHexDigit() && skipHexDigit()"
private boolean skipStringLiteralEscapeSequence(boolean templateLiteral) {
nextChar();
if (isAtEnd()) {
reportError("Unterminated string literal escape sequence");
return false;
}
if (isLineTerminator(peekChar())) {
skipLineTerminator();
return true;
}
char next = nextChar();
switch (next) {
case '\'':
case '"':
case '`':
case '\\':
case 'b':
case 'f':
case 'n':
case 'r':
case 't':
case 'v':
case '0':
return true;
case '1':
case '2':
case '3':
case '4':
case '5':
case '6':
case '7':
if (templateLiteral) {
reportError("Invalid escape sequence");
return false;
}
break;
case 'x':
return skipHexDigit() && skipHexDigit();
case 'u':
if (peek('{')) {
nextChar();
if (peek('}')) {
reportError("Empty unicode escape");
return false;
}
boolean allHexDigits = true;
while (!peek('}') && allHexDigits) {
allHexDigits = allHexDigits && skipHexDigit();
}
nextChar();
return allHexDigits;
} else {
return skipHexDigit() && skipHexDigit() && skipHexDigit() && skipHexDigit();
}
default:
break;
}
if (next == '/') {
// Don't warn for '\/' (for now) since it's common in "<\/script>"
} else if (templateLiteral) {
// Don't warn in template literals since tagged template literals
// can access the raw string value.
} else {
reportWarning("Unnecessary escape: '\\%s' is equivalent to just '%s'", next, next);
}
return true;
}
private boolean skipHexDigit() {
if (!peekHexDigit()) {
reportError("Hex digit expected");
return false;
}
nextChar();
return true;
}
private void skipLineTerminator() {
char first = nextChar();
if (first == '\r' && peek('\n')) {
nextChar();
}
}
private LiteralToken scanFractionalNumericLiteral(int beginToken) {
if (peek('.')) {
nextChar();
skipDecimalDigits();
}
return scanExponentOfNumericLiteral(beginToken);
}
private LiteralToken scanExponentOfNumericLiteral(int beginToken) {
switch (peekChar()) {
case 'e':
case 'E':
nextChar();
switch (peekChar()) {
case '+':
case '-':
nextChar();
break;
default: // fall out
}
if (!isDecimalDigit(peekChar())) {
reportError("Exponent part must contain at least one digit");
}
skipDecimalDigits();
break;
default:
break;
}
return new LiteralToken(
TokenType.NUMBER, getTokenString(beginToken), getTokenRange(beginToken));
}
private void skipDecimalDigits() {
while (isDecimalDigit(peekChar())) {
nextChar();
}
}
private static boolean isDecimalDigit(char ch) {
switch (ch) {
case '0': case '1': case '2': case '3': case '4':
case '5': case '6': case '7': case '8': case '9':
return true;
default:
return false;
}
}
private boolean peekHexDigit() {
return Character.digit(peekChar(), 0x10) >= 0;
}
private void skipHexDigits() {
while (peekHexDigit()) {
nextChar();
}
}
private void skipOctalDigits() {
while (isOctalDigit(peekChar())) {
nextChar();
}
}
private static boolean isOctalDigit(char ch) {
return valueOfOctalDigit(ch) >= 0;
}
private static int valueOfOctalDigit(char ch) {
switch (ch) {
case '0': case '1': case '2': case '3': case '4':
case '5': case '6': case '7':
return ch - '0';
default:
return -1;
}
}
private void skipBinaryDigits() {
while (isBinaryDigit(peekChar())) {
nextChar();
}
}
private static boolean isBinaryDigit(char ch) {
return valueOfBinaryDigit(ch) >= 0;
}
private static int valueOfBinaryDigit(char ch) {
switch (ch) {
case '0':
return 0;
case '1':
return 1;
default:
return -1;
}
}
private char nextChar() {
if (isAtEnd()) {
return '\0';
}
return source.contents.charAt(index++);
}
private boolean peek(char ch) {
return peekChar() == ch;
}
private char peekChar() {
return peekChar(0);
}
private char peekChar(int offset) {
return !isValidIndex(index + offset) ? '\0' : source.contents.charAt(index + offset);
}
@FormatMethod
private void reportError(@FormatString String format, Object... arguments) {
reportError(getPosition(), format, arguments);
}
@FormatMethod
private void reportError(
SourcePosition position, @FormatString String format, Object... arguments) {
errorReporter.reportError(position, format, arguments);
}
@FormatMethod
private void reportWarning(@FormatString String format, Object... arguments) {
errorReporter.reportWarning(getPosition(), format, arguments);
}
void incTypeParameterLevel() {
typeParameterLevel++;
}
void decTypeParameterLevel() {
typeParameterLevel--;
}
}