jdk8u.jaxp.org.apache.xerces.external.impl.XML11EntityScanner Maven / Gradle / Ivy
/*
* Copyright (c) 2015, 2016, Oracle and/or its affiliates. All rights reserved.
*/
/*
* Licensed to the Apache Software Foundation (ASF) under one or more
* contributor license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright ownership.
* The ASF licenses this file to You 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 jdk8u.jaxp.org.apache.xerces.external.impl;
import jdk8u.jaxp.org.apache.xerces.external.impl.XMLScanner.NameType;
import jdk8u.jaxp.org.apache.xerces.external.impl.msg.XMLMessageFormatter;
import jdk8u.jaxp.org.apache.xerces.external.util.XML11Char;
import jdk8u.jaxp.org.apache.xerces.external.util.XMLChar;
import jdk8u.jaxp.org.apache.xerces.external.util.XMLStringBuffer;
import jdk8u.jaxp.org.apache.xerces.external.utils.XMLSecurityManager.Limit;
import jdk8u.jaxp.org.apache.xerces.external.xni.QName;
import jdk8u.jaxp.org.apache.xerces.external.xni.XMLString;
import jdk8u.jaxp.xml.external.stream.Entity;
import java.io.IOException;
/**
* Implements the entity scanner methods in
* the context of XML 1.1.
*
* @xerces.internal
*
* @author Michael Glavassevich, IBM
* @author Neil Graham, IBM
* @version $Id: XML11EntityScanner.java,v 1.5 2010-11-01 04:39:40 joehw Exp $
*/
public class XML11EntityScanner
extends XMLEntityScanner {
//
// Constructors
//
/** Default constructor. */
public XML11EntityScanner() {
super();
} // ()
//
// XMLEntityScanner methods
//
/**
* Returns the next character on the input.
*
* Note: The character is not consumed.
*
* @throws IOException Thrown if i/o error occurs.
* @throws EOFException Thrown on end of file.
*/
public int peekChar() throws IOException {
// load more characters, if needed
if (fCurrentEntity.position == fCurrentEntity.count) {
load(0, true, true);
}
// peek at character
int c = fCurrentEntity.ch[fCurrentEntity.position];
// return peeked character
if (fCurrentEntity.isExternal()) {
return (c != '\r' && c != 0x85 && c != 0x2028) ? c : '\n';
}
else {
return c;
}
} // peekChar():int
/**
* Returns the next character on the input.
*
* Note: The character is consumed.
*
* @throws IOException Thrown if i/o error occurs.
* @throws EOFException Thrown on end of file.
*/
protected int scanChar(NameType nt) throws IOException {
// load more characters, if needed
if (fCurrentEntity.position == fCurrentEntity.count) {
load(0, true, true);
}
// scan character
int offset = fCurrentEntity.position;
int c = fCurrentEntity.ch[fCurrentEntity.position++];
boolean external = false;
if (c == '\n' ||
((c == '\r' || c == 0x85 || c == 0x2028) && (external = fCurrentEntity.isExternal()))) {
fCurrentEntity.lineNumber++;
fCurrentEntity.columnNumber = 1;
if (fCurrentEntity.position == fCurrentEntity.count) {
invokeListeners(1);
fCurrentEntity.ch[0] = (char)c;
load(1, false, false);
offset = 0;
}
if (c == '\r' && external) {
int cc = fCurrentEntity.ch[fCurrentEntity.position++];
if (cc != '\n' && cc != 0x85) {
fCurrentEntity.position--;
}
}
c = '\n';
}
// return character that was scanned
fCurrentEntity.columnNumber++;
if (!detectingVersion) {
checkEntityLimit(nt, fCurrentEntity, offset, fCurrentEntity.position - offset);
}
return c;
} // scanChar():int
/**
* Returns a string matching the NMTOKEN production appearing immediately
* on the input as a symbol, or null if NMTOKEN Name string is present.
*
* Note: The NMTOKEN characters are consumed.
*
* Note: The string returned must be a symbol. The
* SymbolTable can be used for this purpose.
*
* @throws IOException Thrown if i/o error occurs.
* @throws EOFException Thrown on end of file.
*
* @see jdk8u.jaxp.org.apache.xerces.external.util.SymbolTable
* @see jdk8u.jaxp.org.apache.xerces.external.util.XML11Char#isXML11Name
*/
protected String scanNmtoken() throws IOException {
// load more characters, if needed
if (fCurrentEntity.position == fCurrentEntity.count) {
load(0, true, true);
}
// scan nmtoken
int offset = fCurrentEntity.position;
do {
char ch = fCurrentEntity.ch[fCurrentEntity.position];
if (XML11Char.isXML11Name(ch)) {
if (++fCurrentEntity.position == fCurrentEntity.count) {
int length = fCurrentEntity.position - offset;
invokeListeners(length);
if (length == fCurrentEntity.ch.length) {
// bad luck we have to resize our buffer
char[] tmp = new char[fCurrentEntity.ch.length << 1];
System.arraycopy(fCurrentEntity.ch, offset,
tmp, 0, length);
fCurrentEntity.ch = tmp;
}
else {
System.arraycopy(fCurrentEntity.ch, offset,
fCurrentEntity.ch, 0, length);
}
offset = 0;
if (load(length, false, false)) {
break;
}
}
}
else if (XML11Char.isXML11NameHighSurrogate(ch)) {
if (++fCurrentEntity.position == fCurrentEntity.count) {
int length = fCurrentEntity.position - offset;
invokeListeners(length);
if (length == fCurrentEntity.ch.length) {
// bad luck we have to resize our buffer
char[] tmp = new char[fCurrentEntity.ch.length << 1];
System.arraycopy(fCurrentEntity.ch, offset,
tmp, 0, length);
fCurrentEntity.ch = tmp;
}
else {
System.arraycopy(fCurrentEntity.ch, offset,
fCurrentEntity.ch, 0, length);
}
offset = 0;
if (load(length, false, false)) {
--fCurrentEntity.startPosition;
--fCurrentEntity.position;
break;
}
}
char ch2 = fCurrentEntity.ch[fCurrentEntity.position];
if ( !XMLChar.isLowSurrogate(ch2) ||
!XML11Char.isXML11Name(XMLChar.supplemental(ch, ch2)) ) {
--fCurrentEntity.position;
break;
}
if (++fCurrentEntity.position == fCurrentEntity.count) {
int length = fCurrentEntity.position - offset;
invokeListeners(length);
if (length == fCurrentEntity.ch.length) {
// bad luck we have to resize our buffer
char[] tmp = new char[fCurrentEntity.ch.length << 1];
System.arraycopy(fCurrentEntity.ch, offset,
tmp, 0, length);
fCurrentEntity.ch = tmp;
}
else {
System.arraycopy(fCurrentEntity.ch, offset,
fCurrentEntity.ch, 0, length);
}
offset = 0;
if (load(length, false, false)) {
break;
}
}
}
else {
break;
}
}
while (true);
int length = fCurrentEntity.position - offset;
fCurrentEntity.columnNumber += length;
// return nmtoken
String symbol = null;
if (length > 0) {
symbol = fSymbolTable.addSymbol(fCurrentEntity.ch, offset, length);
}
return symbol;
} // scanNmtoken():String
/**
* Returns a string matching the Name production appearing immediately
* on the input as a symbol, or null if no Name string is present.
*
* Note: The Name characters are consumed.
*
* Note: The string returned must be a symbol. The
* SymbolTable can be used for this purpose.
*
* @param nt The type of the name (element or attribute)
*
* @throws IOException Thrown if i/o error occurs.
* @throws EOFException Thrown on end of file.
*
* @see jdk8u.jaxp.org.apache.xerces.external.util.SymbolTable
* @see jdk8u.jaxp.org.apache.xerces.external.util.XML11Char#isXML11Name
* @see jdk8u.jaxp.org.apache.xerces.external.util.XML11Char#isXML11NameStart
*/
protected String scanName(NameType nt) throws IOException {
// load more characters, if needed
if (fCurrentEntity.position == fCurrentEntity.count) {
load(0, true, true);
}
// scan name
int offset = fCurrentEntity.position;
char ch = fCurrentEntity.ch[offset];
if (XML11Char.isXML11NameStart(ch)) {
if (++fCurrentEntity.position == fCurrentEntity.count) {
invokeListeners(1);
fCurrentEntity.ch[0] = ch;
offset = 0;
if (load(1, false, false)) {
fCurrentEntity.columnNumber++;
String symbol = fSymbolTable.addSymbol(fCurrentEntity.ch, 0, 1);
return symbol;
}
}
}
else if (XML11Char.isXML11NameHighSurrogate(ch)) {
if (++fCurrentEntity.position == fCurrentEntity.count) {
invokeListeners(1);
fCurrentEntity.ch[0] = ch;
offset = 0;
if (load(1, false, false)) {
--fCurrentEntity.position;
--fCurrentEntity.startPosition;
return null;
}
}
char ch2 = fCurrentEntity.ch[fCurrentEntity.position];
if ( !XMLChar.isLowSurrogate(ch2) ||
!XML11Char.isXML11NameStart(XMLChar.supplemental(ch, ch2)) ) {
--fCurrentEntity.position;
return null;
}
if (++fCurrentEntity.position == fCurrentEntity.count) {
invokeListeners(2);
fCurrentEntity.ch[0] = ch;
fCurrentEntity.ch[1] = ch2;
offset = 0;
if (load(2, false, false)) {
fCurrentEntity.columnNumber += 2;
String symbol = fSymbolTable.addSymbol(fCurrentEntity.ch, 0, 2);
return symbol;
}
}
}
else {
return null;
}
int length = 0;
do {
ch = fCurrentEntity.ch[fCurrentEntity.position];
if (XML11Char.isXML11Name(ch)) {
if ((length = checkBeforeLoad(fCurrentEntity, offset, offset)) > 0) {
offset = 0;
if (load(length, false, false)) {
break;
}
}
}
else if (XML11Char.isXML11NameHighSurrogate(ch)) {
if ((length = checkBeforeLoad(fCurrentEntity, offset, offset)) > 0) {
offset = 0;
if (load(length, false, false)) {
--fCurrentEntity.position;
--fCurrentEntity.startPosition;
break;
}
}
char ch2 = fCurrentEntity.ch[fCurrentEntity.position];
if ( !XMLChar.isLowSurrogate(ch2) ||
!XML11Char.isXML11Name(XMLChar.supplemental(ch, ch2)) ) {
--fCurrentEntity.position;
break;
}
if ((length = checkBeforeLoad(fCurrentEntity, offset, offset)) > 0) {
offset = 0;
if (load(length, false, false)) {
break;
}
}
}
else {
break;
}
}
while (true);
length = fCurrentEntity.position - offset;
fCurrentEntity.columnNumber += length;
// return name
String symbol = null;
if (length > 0) {
checkLimit(Limit.MAX_NAME_LIMIT, fCurrentEntity, offset, length);
checkEntityLimit(nt, fCurrentEntity, offset, length);
symbol = fSymbolTable.addSymbol(fCurrentEntity.ch, offset, length);
}
return symbol;
} // scanName():String
/**
* Returns a string matching the NCName production appearing immediately
* on the input as a symbol, or null if no NCName string is present.
*
* Note: The NCName characters are consumed.
*
* Note: The string returned must be a symbol. The
* SymbolTable can be used for this purpose.
*
* @throws IOException Thrown if i/o error occurs.
* @throws EOFException Thrown on end of file.
*
* @see jdk8u.jaxp.org.apache.xerces.external.util.SymbolTable
* @see jdk8u.jaxp.org.apache.xerces.external.util.XML11Char#isXML11NCName
* @see jdk8u.jaxp.org.apache.xerces.external.util.XML11Char#isXML11NCNameStart
*/
protected String scanNCName() throws IOException {
// load more characters, if needed
if (fCurrentEntity.position == fCurrentEntity.count) {
load(0, true, true);
}
// scan name
int offset = fCurrentEntity.position;
char ch = fCurrentEntity.ch[offset];
if (XML11Char.isXML11NCNameStart(ch)) {
if (++fCurrentEntity.position == fCurrentEntity.count) {
invokeListeners(1);
fCurrentEntity.ch[0] = ch;
offset = 0;
if (load(1, false, false)) {
fCurrentEntity.columnNumber++;
String symbol = fSymbolTable.addSymbol(fCurrentEntity.ch, 0, 1);
return symbol;
}
}
}
else if (XML11Char.isXML11NameHighSurrogate(ch)) {
if (++fCurrentEntity.position == fCurrentEntity.count) {
invokeListeners(1);
fCurrentEntity.ch[0] = ch;
offset = 0;
if (load(1, false, false)) {
--fCurrentEntity.position;
--fCurrentEntity.startPosition;
return null;
}
}
char ch2 = fCurrentEntity.ch[fCurrentEntity.position];
if ( !XMLChar.isLowSurrogate(ch2) ||
!XML11Char.isXML11NCNameStart(XMLChar.supplemental(ch, ch2)) ) {
--fCurrentEntity.position;
return null;
}
if (++fCurrentEntity.position == fCurrentEntity.count) {
invokeListeners(2);
fCurrentEntity.ch[0] = ch;
fCurrentEntity.ch[1] = ch2;
offset = 0;
if (load(2, false, false)) {
fCurrentEntity.columnNumber += 2;
String symbol = fSymbolTable.addSymbol(fCurrentEntity.ch, 0, 2);
return symbol;
}
}
}
else {
return null;
}
do {
ch = fCurrentEntity.ch[fCurrentEntity.position];
if (XML11Char.isXML11NCName(ch)) {
if (++fCurrentEntity.position == fCurrentEntity.count) {
int length = fCurrentEntity.position - offset;
invokeListeners(length);
if (length == fCurrentEntity.ch.length) {
// bad luck we have to resize our buffer
char[] tmp = new char[fCurrentEntity.ch.length << 1];
System.arraycopy(fCurrentEntity.ch, offset,
tmp, 0, length);
fCurrentEntity.ch = tmp;
}
else {
System.arraycopy(fCurrentEntity.ch, offset,
fCurrentEntity.ch, 0, length);
}
offset = 0;
if (load(length, false, false)) {
break;
}
}
}
else if (XML11Char.isXML11NameHighSurrogate(ch)) {
if (++fCurrentEntity.position == fCurrentEntity.count) {
int length = fCurrentEntity.position - offset;
invokeListeners(length);
if (length == fCurrentEntity.ch.length) {
// bad luck we have to resize our buffer
char[] tmp = new char[fCurrentEntity.ch.length << 1];
System.arraycopy(fCurrentEntity.ch, offset,
tmp, 0, length);
fCurrentEntity.ch = tmp;
}
else {
System.arraycopy(fCurrentEntity.ch, offset,
fCurrentEntity.ch, 0, length);
}
offset = 0;
if (load(length, false, false)) {
--fCurrentEntity.startPosition;
--fCurrentEntity.position;
break;
}
}
char ch2 = fCurrentEntity.ch[fCurrentEntity.position];
if ( !XMLChar.isLowSurrogate(ch2) ||
!XML11Char.isXML11NCName(XMLChar.supplemental(ch, ch2)) ) {
--fCurrentEntity.position;
break;
}
if (++fCurrentEntity.position == fCurrentEntity.count) {
int length = fCurrentEntity.position - offset;
invokeListeners(length);
if (length == fCurrentEntity.ch.length) {
// bad luck we have to resize our buffer
char[] tmp = new char[fCurrentEntity.ch.length << 1];
System.arraycopy(fCurrentEntity.ch, offset,
tmp, 0, length);
fCurrentEntity.ch = tmp;
}
else {
System.arraycopy(fCurrentEntity.ch, offset,
fCurrentEntity.ch, 0, length);
}
offset = 0;
if (load(length, false, false)) {
break;
}
}
}
else {
break;
}
}
while (true);
int length = fCurrentEntity.position - offset;
fCurrentEntity.columnNumber += length;
// return name
String symbol = null;
if (length > 0) {
symbol = fSymbolTable.addSymbol(fCurrentEntity.ch, offset, length);
}
return symbol;
} // scanNCName():String
/**
* Scans a qualified name from the input, setting the fields of the
* QName structure appropriately.
*
* Note: The qualified name characters are consumed.
*
* Note: The strings used to set the values of the
* QName structure must be symbols. The SymbolTable can be used for
* this purpose.
*
* @param qname The qualified name structure to fill.
* @param nt The type of the name (element or attribute)
*
* @return Returns true if a qualified name appeared immediately on
* the input and was scanned, false otherwise.
*
* @throws IOException Thrown if i/o error occurs.
* @throws EOFException Thrown on end of file.
*
* @see jdk8u.jaxp.org.apache.xerces.external.util.SymbolTable
* @see jdk8u.jaxp.org.apache.xerces.external.util.XML11Char#isXML11Name
* @see jdk8u.jaxp.org.apache.xerces.external.util.XML11Char#isXML11NameStart
*/
protected boolean scanQName(QName qname, XMLScanner.NameType nt) throws IOException {
// load more characters, if needed
if (fCurrentEntity.position == fCurrentEntity.count) {
load(0, true, true);
}
// scan qualified name
int offset = fCurrentEntity.position;
char ch = fCurrentEntity.ch[offset];
if (XML11Char.isXML11NCNameStart(ch)) {
if (++fCurrentEntity.position == fCurrentEntity.count) {
invokeListeners(1);
fCurrentEntity.ch[0] = ch;
offset = 0;
if (load(1, false, false)) {
fCurrentEntity.columnNumber++;
String name = fSymbolTable.addSymbol(fCurrentEntity.ch, 0, 1);
qname.setValues(null, name, name, null);
checkEntityLimit(nt, fCurrentEntity, 0, 1);
return true;
}
}
}
else if (XML11Char.isXML11NameHighSurrogate(ch)) {
if (++fCurrentEntity.position == fCurrentEntity.count) {
invokeListeners(1);
fCurrentEntity.ch[0] = ch;
offset = 0;
if (load(1, false, false)) {
--fCurrentEntity.startPosition;
--fCurrentEntity.position;
return false;
}
}
char ch2 = fCurrentEntity.ch[fCurrentEntity.position];
if ( !XMLChar.isLowSurrogate(ch2) ||
!XML11Char.isXML11NCNameStart(XMLChar.supplemental(ch, ch2)) ) {
--fCurrentEntity.position;
return false;
}
if (++fCurrentEntity.position == fCurrentEntity.count) {
invokeListeners(2);
fCurrentEntity.ch[0] = ch;
fCurrentEntity.ch[1] = ch2;
offset = 0;
if (load(2, false, false)) {
fCurrentEntity.columnNumber += 2;
String name = fSymbolTable.addSymbol(fCurrentEntity.ch, 0, 2);
qname.setValues(null, name, name, null);
checkEntityLimit(nt, fCurrentEntity, 0, 2);
return true;
}
}
}
else {
return false;
}
int index = -1;
int length = 0;
boolean sawIncompleteSurrogatePair = false;
do {
ch = fCurrentEntity.ch[fCurrentEntity.position];
if (XML11Char.isXML11Name(ch)) {
if (ch == ':') {
if (index != -1) {
break;
}
index = fCurrentEntity.position;
//check prefix before further read
checkLimit(Limit.MAX_NAME_LIMIT, fCurrentEntity, offset, index - offset);
}
if ((length = checkBeforeLoad(fCurrentEntity, offset, index)) > 0) {
if (index != -1) {
index = index - offset;
}
offset = 0;
if (load(length, false, false)) {
break;
}
}
}
else if (XML11Char.isXML11NameHighSurrogate(ch)) {
if ((length = checkBeforeLoad(fCurrentEntity, offset, index)) > 0) {
if (index != -1) {
index = index - offset;
}
offset = 0;
if (load(length, false, false)) {
sawIncompleteSurrogatePair = true;
--fCurrentEntity.startPosition;
--fCurrentEntity.position;
break;
}
}
char ch2 = fCurrentEntity.ch[fCurrentEntity.position];
if ( !XMLChar.isLowSurrogate(ch2) ||
!XML11Char.isXML11Name(XMLChar.supplemental(ch, ch2)) ) {
sawIncompleteSurrogatePair = true;
--fCurrentEntity.position;
break;
}
if ((length = checkBeforeLoad(fCurrentEntity, offset, index)) > 0) {
if (index != -1) {
index = index - offset;
}
offset = 0;
if (load(length, false, false)) {
break;
}
}
}
else {
break;
}
}
while (true);
length = fCurrentEntity.position - offset;
fCurrentEntity.columnNumber += length;
if (length > 0) {
String prefix = null;
String localpart = null;
String rawname = fSymbolTable.addSymbol(fCurrentEntity.ch,
offset, length);
if (index != -1) {
int prefixLength = index - offset;
//check the result: prefix
checkLimit(Limit.MAX_NAME_LIMIT, fCurrentEntity, offset, prefixLength);
prefix = fSymbolTable.addSymbol(fCurrentEntity.ch,
offset, prefixLength);
int len = length - prefixLength - 1;
int startLocal = index +1;
if (!XML11Char.isXML11NCNameStart(fCurrentEntity.ch[startLocal]) &&
(!XML11Char.isXML11NameHighSurrogate(fCurrentEntity.ch[startLocal]) ||
sawIncompleteSurrogatePair)){
fErrorReporter.reportError(XMLMessageFormatter.XML_DOMAIN,
"IllegalQName",
null,
XMLErrorReporter.SEVERITY_FATAL_ERROR);
}
//check the result: localpart
checkLimit(Limit.MAX_NAME_LIMIT, fCurrentEntity, index + 1, len);
localpart = fSymbolTable.addSymbol(fCurrentEntity.ch,
index + 1, len);
}
else {
localpart = rawname;
//check the result: localpart
checkLimit(Limit.MAX_NAME_LIMIT, fCurrentEntity, offset, length);
}
qname.setValues(prefix, localpart, rawname, null);
checkEntityLimit(nt, fCurrentEntity, offset, length);
return true;
}
return false;
} // scanQName(QName):boolean
/**
* Scans a range of parsed character data, setting the fields of the
* XMLString structure, appropriately.
*
* Note: The characters are consumed.
*
* Note: This method does not guarantee to return
* the longest run of parsed character data. This method may return
* before markup due to reaching the end of the input buffer or any
* other reason.
*
* Note: The fields contained in the XMLString
* structure are not guaranteed to remain valid upon subsequent calls
* to the entity scanner. Therefore, the caller is responsible for
* immediately using the returned character data or making a copy of
* the character data.
*
* @param content The content structure to fill.
*
* @return Returns the next character on the input, if known. This
* value may be -1 but this does note designate
* end of file.
*
* @throws IOException Thrown if i/o error occurs.
* @throws EOFException Thrown on end of file.
*/
protected int scanContent(XMLString content) throws IOException {
// load more characters, if needed
if (fCurrentEntity.position == fCurrentEntity.count) {
load(0, true, true);
}
else if (fCurrentEntity.position == fCurrentEntity.count - 1) {
invokeListeners(1);
fCurrentEntity.ch[0] = fCurrentEntity.ch[fCurrentEntity.count - 1];
load(1, false, false);
fCurrentEntity.position = 0;
fCurrentEntity.startPosition = 0;
}
// normalize newlines
int offset = fCurrentEntity.position;
int c = fCurrentEntity.ch[offset];
int newlines = 0;
boolean counted = false;
boolean external = fCurrentEntity.isExternal();
if (c == '\n' || ((c == '\r' || c == 0x85 || c == 0x2028) && external)) {
do {
c = fCurrentEntity.ch[fCurrentEntity.position++];
if ((c == '\r' ) && external) {
newlines++;
fCurrentEntity.lineNumber++;
fCurrentEntity.columnNumber = 1;
if (fCurrentEntity.position == fCurrentEntity.count) {
checkEntityLimit(null, fCurrentEntity, offset, newlines);
offset = 0;
fCurrentEntity.baseCharOffset += (fCurrentEntity.position - fCurrentEntity.startPosition);
fCurrentEntity.position = newlines;
fCurrentEntity.startPosition = newlines;
if (load(newlines, false, true)) {
counted = true;
break;
}
}
int cc = fCurrentEntity.ch[fCurrentEntity.position];
if (cc == '\n' || cc == 0x85) {
fCurrentEntity.position++;
offset++;
}
/*** NEWLINE NORMALIZATION ***/
else {
newlines++;
}
}
else if (c == '\n' || ((c == 0x85 || c == 0x2028) && external)) {
newlines++;
fCurrentEntity.lineNumber++;
fCurrentEntity.columnNumber = 1;
if (fCurrentEntity.position == fCurrentEntity.count) {
checkEntityLimit(null, fCurrentEntity, offset, newlines);
offset = 0;
fCurrentEntity.baseCharOffset += (fCurrentEntity.position - fCurrentEntity.startPosition);
fCurrentEntity.position = newlines;
fCurrentEntity.startPosition = newlines;
if (load(newlines, false, true)) {
counted = true;
break;
}
}
}
else {
fCurrentEntity.position--;
break;
}
} while (fCurrentEntity.position < fCurrentEntity.count - 1);
for (int i = offset; i < fCurrentEntity.position; i++) {
fCurrentEntity.ch[i] = '\n';
}
int length = fCurrentEntity.position - offset;
if (fCurrentEntity.position == fCurrentEntity.count - 1) {
checkEntityLimit(null, fCurrentEntity, offset, length);
content.setValues(fCurrentEntity.ch, offset, length);
return -1;
}
}
// inner loop, scanning for content
if (external) {
while (fCurrentEntity.position < fCurrentEntity.count) {
c = fCurrentEntity.ch[fCurrentEntity.position++];
if (!XML11Char.isXML11Content(c) || c == 0x85 || c == 0x2028) {
fCurrentEntity.position--;
break;
}
}
}
else {
while (fCurrentEntity.position < fCurrentEntity.count) {
c = fCurrentEntity.ch[fCurrentEntity.position++];
// In internal entities control characters are allowed to appear unescaped.
if (!XML11Char.isXML11InternalEntityContent(c)) {
fCurrentEntity.position--;
break;
}
}
}
int length = fCurrentEntity.position - offset;
fCurrentEntity.columnNumber += length - newlines;
if (!counted) {
checkEntityLimit(null, fCurrentEntity, offset, length);
}
content.setValues(fCurrentEntity.ch, offset, length);
// return next character
if (fCurrentEntity.position != fCurrentEntity.count) {
c = fCurrentEntity.ch[fCurrentEntity.position];
// REVISIT: Does this need to be updated to fix the
// #x0D ^#x0A newline normalization problem? -Ac
if ((c == '\r' || c == 0x85 || c == 0x2028) && external) {
c = '\n';
}
}
else {
c = -1;
}
return c;
} // scanContent(XMLString):int
/**
* Scans a range of attribute value data, setting the fields of the
* XMLString structure, appropriately.
*
* Note: The characters are consumed.
*
* Note: This method does not guarantee to return
* the longest run of attribute value data. This method may return
* before the quote character due to reaching the end of the input
* buffer or any other reason.
*
* Note: The fields contained in the XMLString
* structure are not guaranteed to remain valid upon subsequent calls
* to the entity scanner. Therefore, the caller is responsible for
* immediately using the returned character data or making a copy of
* the character data.
*
* @param quote The quote character that signifies the end of the
* attribute value data.
* @param content The content structure to fill.
* @param isNSURI a flag indicating whether the content is a Namespace URI
*
* @return Returns the next character on the input, if known. This
* value may be -1 but this does note designate
* end of file.
*
* @throws IOException Thrown if i/o error occurs.
* @throws EOFException Thrown on end of file.
*/
protected int scanLiteral(int quote, XMLString content, boolean isNSURI)
throws IOException {
// load more characters, if needed
if (fCurrentEntity.position == fCurrentEntity.count) {
load(0, true, true);
}
else if (fCurrentEntity.position == fCurrentEntity.count - 1) {
invokeListeners(1);
fCurrentEntity.ch[0] = fCurrentEntity.ch[fCurrentEntity.count - 1];
load(1, false, false);
fCurrentEntity.startPosition = 0;
fCurrentEntity.position = 0;
}
// normalize newlines
int offset = fCurrentEntity.position;
int c = fCurrentEntity.ch[offset];
int newlines = 0;
boolean external = fCurrentEntity.isExternal();
if (c == '\n' || ((c == '\r' || c == 0x85 || c == 0x2028) && external)) {
do {
c = fCurrentEntity.ch[fCurrentEntity.position++];
if ((c == '\r' ) && external) {
newlines++;
fCurrentEntity.lineNumber++;
fCurrentEntity.columnNumber = 1;
if (fCurrentEntity.position == fCurrentEntity.count) {
offset = 0;
fCurrentEntity.baseCharOffset += (fCurrentEntity.position - fCurrentEntity.startPosition);
fCurrentEntity.position = newlines;
fCurrentEntity.startPosition = newlines;
if (load(newlines, false, true)) {
break;
}
}
int cc = fCurrentEntity.ch[fCurrentEntity.position];
if (cc == '\n' || cc == 0x85) {
fCurrentEntity.position++;
offset++;
}
/*** NEWLINE NORMALIZATION ***/
else {
newlines++;
}
}
else if (c == '\n' || ((c == 0x85 || c == 0x2028) && external)) {
newlines++;
fCurrentEntity.lineNumber++;
fCurrentEntity.columnNumber = 1;
if (fCurrentEntity.position == fCurrentEntity.count) {
offset = 0;
fCurrentEntity.baseCharOffset += (fCurrentEntity.position - fCurrentEntity.startPosition);
fCurrentEntity.position = newlines;
fCurrentEntity.startPosition = newlines;
if (load(newlines, false, true)) {
break;
}
}
}
else {
fCurrentEntity.position--;
break;
}
} while (fCurrentEntity.position < fCurrentEntity.count - 1);
for (int i = offset; i < fCurrentEntity.position; i++) {
fCurrentEntity.ch[i] = '\n';
}
int length = fCurrentEntity.position - offset;
if (fCurrentEntity.position == fCurrentEntity.count - 1) {
content.setValues(fCurrentEntity.ch, offset, length);
return -1;
}
}
// scan literal value
if (external) {
while (fCurrentEntity.position < fCurrentEntity.count) {
c = fCurrentEntity.ch[fCurrentEntity.position++];
if (c == quote || c == '%' || !XML11Char.isXML11Content(c)
|| c == 0x85 || c == 0x2028) {
fCurrentEntity.position--;
break;
}
}
}
else {
while (fCurrentEntity.position < fCurrentEntity.count) {
c = fCurrentEntity.ch[fCurrentEntity.position++];
// In internal entities control characters are allowed to appear unescaped.
if ((c == quote && !fCurrentEntity.literal)
|| c == '%' || !XML11Char.isXML11InternalEntityContent(c)) {
fCurrentEntity.position--;
break;
}
}
}
int length = fCurrentEntity.position - offset;
fCurrentEntity.columnNumber += length - newlines;
checkEntityLimit(null, fCurrentEntity, offset, length);
if (isNSURI) {
checkLimit(Limit.MAX_NAME_LIMIT, fCurrentEntity, offset, length);
}
content.setValues(fCurrentEntity.ch, offset, length);
// return next character
if (fCurrentEntity.position != fCurrentEntity.count) {
c = fCurrentEntity.ch[fCurrentEntity.position];
// NOTE: We don't want to accidentally signal the
// end of the literal if we're expanding an
// entity appearing in the literal. -Ac
if (c == quote && fCurrentEntity.literal) {
c = -1;
}
}
else {
c = -1;
}
return c;
} // scanLiteral(int,XMLString):int
/**
* Scans a range of character data up to the specicied delimiter,
* setting the fields of the XMLString structure, appropriately.
*
* Note: The characters are consumed.
*
* Note: This assumes that the internal buffer is
* at least the same size, or bigger, than the length of the delimiter
* and that the delimiter contains at least one character.
*
* Note: This method does not guarantee to return
* the longest run of character data. This method may return before
* the delimiter due to reaching the end of the input buffer or any
* other reason.
*
* Note: The fields contained in the XMLString
* structure are not guaranteed to remain valid upon subsequent calls
* to the entity scanner. Therefore, the caller is responsible for
* immediately using the returned character data or making a copy of
* the character data.
*
* @param delimiter The string that signifies the end of the character
* data to be scanned.
* @param data The data structure to fill.
*
* @return Returns true if there is more data to scan, false otherwise.
*
* @throws IOException Thrown if i/o error occurs.
* @throws EOFException Thrown on end of file.
*/
protected boolean scanData(String delimiter, XMLStringBuffer buffer)
throws IOException {
boolean done = false;
int delimLen = delimiter.length();
char charAt0 = delimiter.charAt(0);
boolean external = fCurrentEntity.isExternal();
do {
// load more characters, if needed
if (fCurrentEntity.position == fCurrentEntity.count) {
load(0, true, false);
}
boolean bNextEntity = false;
while ((fCurrentEntity.position >= fCurrentEntity.count - delimLen)
&& (!bNextEntity))
{
System.arraycopy(fCurrentEntity.ch,
fCurrentEntity.position,
fCurrentEntity.ch,
0,
fCurrentEntity.count - fCurrentEntity.position);
bNextEntity = load(fCurrentEntity.count - fCurrentEntity.position, false, false);
fCurrentEntity.position = 0;
fCurrentEntity.startPosition = 0;
}
if (fCurrentEntity.position >= fCurrentEntity.count - delimLen) {
// something must be wrong with the input: e.g., file ends an unterminated comment
int length = fCurrentEntity.count - fCurrentEntity.position;
checkEntityLimit(NameType.COMMENT, fCurrentEntity, fCurrentEntity.position, length);
buffer.append (fCurrentEntity.ch, fCurrentEntity.position, length);
fCurrentEntity.columnNumber += fCurrentEntity.count;
fCurrentEntity.baseCharOffset += (fCurrentEntity.position - fCurrentEntity.startPosition);
fCurrentEntity.position = fCurrentEntity.count;
fCurrentEntity.startPosition = fCurrentEntity.count;
load(0,true, false);
return false;
}
// normalize newlines
int offset = fCurrentEntity.position;
int c = fCurrentEntity.ch[offset];
int newlines = 0;
if (c == '\n' || ((c == '\r' || c == 0x85 || c == 0x2028) && external)) {
do {
c = fCurrentEntity.ch[fCurrentEntity.position++];
if ((c == '\r' ) && external) {
newlines++;
fCurrentEntity.lineNumber++;
fCurrentEntity.columnNumber = 1;
if (fCurrentEntity.position == fCurrentEntity.count) {
offset = 0;
fCurrentEntity.baseCharOffset += (fCurrentEntity.position - fCurrentEntity.startPosition);
fCurrentEntity.position = newlines;
fCurrentEntity.startPosition = newlines;
if (load(newlines, false, true)) {
break;
}
}
int cc = fCurrentEntity.ch[fCurrentEntity.position];
if (cc == '\n' || cc == 0x85) {
fCurrentEntity.position++;
offset++;
}
/*** NEWLINE NORMALIZATION ***/
else {
newlines++;
}
}
else if (c == '\n' || ((c == 0x85 || c == 0x2028) && external)) {
newlines++;
fCurrentEntity.lineNumber++;
fCurrentEntity.columnNumber = 1;
if (fCurrentEntity.position == fCurrentEntity.count) {
offset = 0;
fCurrentEntity.baseCharOffset += (fCurrentEntity.position - fCurrentEntity.startPosition);
fCurrentEntity.position = newlines;
fCurrentEntity.startPosition = newlines;
fCurrentEntity.count = newlines;
if (load(newlines, false, true)) {
break;
}
}
}
else {
fCurrentEntity.position--;
break;
}
} while (fCurrentEntity.position < fCurrentEntity.count - 1);
for (int i = offset; i < fCurrentEntity.position; i++) {
fCurrentEntity.ch[i] = '\n';
}
int length = fCurrentEntity.position - offset;
if (fCurrentEntity.position == fCurrentEntity.count - 1) {
checkEntityLimit(NameType.COMMENT, fCurrentEntity, offset, length);
buffer.append(fCurrentEntity.ch, offset, length);
return true;
}
}
// iterate over buffer looking for delimiter
if (external) {
OUTER: while (fCurrentEntity.position < fCurrentEntity.count) {
c = fCurrentEntity.ch[fCurrentEntity.position++];
if (c == charAt0) {
// looks like we just hit the delimiter
int delimOffset = fCurrentEntity.position - 1;
for (int i = 1; i < delimLen; i++) {
if (fCurrentEntity.position == fCurrentEntity.count) {
fCurrentEntity.position -= i;
break OUTER;
}
c = fCurrentEntity.ch[fCurrentEntity.position++];
if (delimiter.charAt(i) != c) {
fCurrentEntity.position--;
break;
}
}
if (fCurrentEntity.position == delimOffset + delimLen) {
done = true;
break;
}
}
else if (c == '\n' || c == '\r' || c == 0x85 || c == 0x2028) {
fCurrentEntity.position--;
break;
}
// In external entities control characters cannot appear
// as literals so do not skip over them.
else if (!XML11Char.isXML11ValidLiteral(c)) {
fCurrentEntity.position--;
int length = fCurrentEntity.position - offset;
fCurrentEntity.columnNumber += length - newlines;
checkEntityLimit(NameType.COMMENT, fCurrentEntity, offset, length);
buffer.append(fCurrentEntity.ch, offset, length);
return true;
}
}
}
else {
OUTER: while (fCurrentEntity.position < fCurrentEntity.count) {
c = fCurrentEntity.ch[fCurrentEntity.position++];
if (c == charAt0) {
// looks like we just hit the delimiter
int delimOffset = fCurrentEntity.position - 1;
for (int i = 1; i < delimLen; i++) {
if (fCurrentEntity.position == fCurrentEntity.count) {
fCurrentEntity.position -= i;
break OUTER;
}
c = fCurrentEntity.ch[fCurrentEntity.position++];
if (delimiter.charAt(i) != c) {
fCurrentEntity.position--;
break;
}
}
if (fCurrentEntity.position == delimOffset + delimLen) {
done = true;
break;
}
}
else if (c == '\n') {
fCurrentEntity.position--;
break;
}
// Control characters are allowed to appear as literals
// in internal entities.
else if (!XML11Char.isXML11Valid(c)) {
fCurrentEntity.position--;
int length = fCurrentEntity.position - offset;
fCurrentEntity.columnNumber += length - newlines;
checkEntityLimit(NameType.COMMENT, fCurrentEntity, offset, length);
buffer.append(fCurrentEntity.ch, offset, length);
return true;
}
}
}
int length = fCurrentEntity.position - offset;
fCurrentEntity.columnNumber += length - newlines;
checkEntityLimit(NameType.COMMENT, fCurrentEntity, offset, length);
if (done) {
length -= delimLen;
}
buffer.append(fCurrentEntity.ch, offset, length);
// return true if string was skipped
} while (!done);
return !done;
} // scanData(String,XMLString)
/**
* Skips a character appearing immediately on the input.
*
* Note: The character is consumed only if it matches
* the specified character.
*
* @param c The character to skip.
*
* @return Returns true if the character was skipped.
*
* @throws IOException Thrown if i/o error occurs.
* @throws EOFException Thrown on end of file.
*/
protected boolean skipChar(int c, NameType nt) throws IOException {
// load more characters, if needed
if (fCurrentEntity.position == fCurrentEntity.count) {
load(0, true, true);
}
// skip character
int offset = fCurrentEntity.position;
int cc = fCurrentEntity.ch[fCurrentEntity.position];
if (cc == c) {
fCurrentEntity.position++;
if (c == '\n') {
fCurrentEntity.lineNumber++;
fCurrentEntity.columnNumber = 1;
}
else {
fCurrentEntity.columnNumber++;
}
checkEntityLimit(nt, fCurrentEntity, offset, fCurrentEntity.position - offset);
return true;
}
else if (c == '\n' && ((cc == 0x2028 || cc == 0x85) && fCurrentEntity.isExternal())) {
fCurrentEntity.position++;
fCurrentEntity.lineNumber++;
fCurrentEntity.columnNumber = 1;
checkEntityLimit(nt, fCurrentEntity, offset, fCurrentEntity.position - offset);
return true;
}
else if (c == '\n' && (cc == '\r' ) && fCurrentEntity.isExternal()) {
// handle newlines
if (fCurrentEntity.position == fCurrentEntity.count) {
invokeListeners(1);
fCurrentEntity.ch[0] = (char)cc;
load(1, false, false);
}
int ccc = fCurrentEntity.ch[++fCurrentEntity.position];
if (ccc == '\n' || ccc == 0x85) {
fCurrentEntity.position++;
}
fCurrentEntity.lineNumber++;
fCurrentEntity.columnNumber = 1;
checkEntityLimit(nt, fCurrentEntity, offset, fCurrentEntity.position - offset);
return true;
}
// character was not skipped
return false;
} // skipChar(int):boolean
/**
* Skips space characters appearing immediately on the input.
*
* Note: The characters are consumed only if they are
* space characters.
*
* @return Returns true if at least one space character was skipped.
*
* @throws IOException Thrown if i/o error occurs.
* @throws EOFException Thrown on end of file.
*
* @see jdk8u.jaxp.org.apache.xerces.external.util.XMLChar#isSpace
* @see jdk8u.jaxp.org.apache.xerces.external.util.XML11Char#isXML11Space
*/
protected boolean skipSpaces() throws IOException {
// load more characters, if needed
if (fCurrentEntity.position == fCurrentEntity.count) {
load(0, true, true);
}
//we are doing this check only in skipSpace() because it is called by
//fMiscDispatcher and we want the parser to exit gracefully when document
//is well-formed.
//it is possible that end of document is reached and
//fCurrentEntity becomes null
//nothing was read so entity changed 'false' should be returned.
if(fCurrentEntity == null){
return false ;
}
// skip spaces
int c = fCurrentEntity.ch[fCurrentEntity.position];
int offset = fCurrentEntity.position - 1;
// External -- Match: S + 0x85 + 0x2028, and perform end of line normalization
if (fCurrentEntity.isExternal()) {
if (XML11Char.isXML11Space(c)) {
do {
boolean entityChanged = false;
// handle newlines
if (c == '\n' || c == '\r' || c == 0x85 || c == 0x2028) {
fCurrentEntity.lineNumber++;
fCurrentEntity.columnNumber = 1;
if (fCurrentEntity.position == fCurrentEntity.count - 1) {
invokeListeners(1);
fCurrentEntity.ch[0] = (char)c;
entityChanged = load(1, true, false);
if (!entityChanged) {
// the load change the position to be 1,
// need to restore it when entity not changed
fCurrentEntity.startPosition = 0;
fCurrentEntity.position = 0;
} else if(fCurrentEntity == null){
return true ;
}
}
if (c == '\r') {
// REVISIT: Does this need to be updated to fix the
// #x0D ^#x0A newline normalization problem? -Ac
int cc = fCurrentEntity.ch[++fCurrentEntity.position];
if (cc != '\n' && cc != 0x85 ) {
fCurrentEntity.position--;
}
}
}
else {
fCurrentEntity.columnNumber++;
}
//If this is a general entity, spaces within a start element should be counted
checkEntityLimit(null, fCurrentEntity, offset, fCurrentEntity.position - offset);
offset = fCurrentEntity.position;
// load more characters, if needed
if (!entityChanged)
fCurrentEntity.position++;
if (fCurrentEntity.position == fCurrentEntity.count) {
load(0, true, true);
if(fCurrentEntity == null){
return true ;
}
}
} while (XML11Char.isXML11Space(c = fCurrentEntity.ch[fCurrentEntity.position]));
return true;
}
}
// Internal -- Match: S (only)
else if (XMLChar.isSpace(c)) {
do {
boolean entityChanged = false;
// handle newlines
if (c == '\n') {
fCurrentEntity.lineNumber++;
fCurrentEntity.columnNumber = 1;
if (fCurrentEntity.position == fCurrentEntity.count - 1) {
invokeListeners(1);
fCurrentEntity.ch[0] = (char)c;
entityChanged = load(1, true, false);
if (!entityChanged) {
// the load change the position to be 1,
// need to restore it when entity not changed
fCurrentEntity.startPosition = 0;
fCurrentEntity.position = 0;
} else if(fCurrentEntity == null){
return true ;
}
}
}
else {
fCurrentEntity.columnNumber++;
}
//If this is a general entity, spaces within a start element should be counted
checkEntityLimit(null, fCurrentEntity, offset, fCurrentEntity.position - offset);
offset = fCurrentEntity.position;
// load more characters, if needed
if (!entityChanged)
fCurrentEntity.position++;
if (fCurrentEntity.position == fCurrentEntity.count) {
load(0, true, true);
if(fCurrentEntity == null){
return true ;
}
}
} while (XMLChar.isSpace(c = fCurrentEntity.ch[fCurrentEntity.position]));
return true;
}
// no spaces were found
return false;
} // skipSpaces():boolean
/**
* Skips the specified string appearing immediately on the input.
*
* Note: The characters are consumed only if they are
* space characters.
*
* @param s The string to skip.
*
* @return Returns true if the string was skipped.
*
* @throws IOException Thrown if i/o error occurs.
* @throws EOFException Thrown on end of file.
*/
protected boolean skipString(String s) throws IOException {
// load more characters, if needed
if (fCurrentEntity.position == fCurrentEntity.count) {
load(0, true, true);
}
// skip string
final int length = s.length();
final int beforeSkip = fCurrentEntity.position ;
for (int i = 0; i < length; i++) {
char c = fCurrentEntity.ch[fCurrentEntity.position++];
if (c != s.charAt(i)) {
fCurrentEntity.position -= i + 1;
return false;
}
if (i < length - 1 && fCurrentEntity.position == fCurrentEntity.count) {
invokeListeners(0);
System.arraycopy(fCurrentEntity.ch, fCurrentEntity.count - i - 1, fCurrentEntity.ch, 0, i + 1);
// REVISIT: Can a string to be skipped cross an
// entity boundary? -Ac
if (load(i + 1, false, false)) {
fCurrentEntity.startPosition -= i + 1;
fCurrentEntity.position -= i + 1;
return false;
}
}
}
fCurrentEntity.columnNumber += length;
if (!detectingVersion) {
checkEntityLimit(null, fCurrentEntity, beforeSkip, length);
}
return true;
} // skipString(String):boolean
} // class XML11EntityScanner