All Downloads are FREE. Search and download functionalities are using the official Maven repository.

src.java.net.htmlparser.jericho.StreamedSource Maven / Gradle / Ivy

There is a newer version: 5.0.84
Show newest version
// Jericho HTML Parser - Java based library for analysing and manipulating HTML
// Version 3.1
// Copyright (C) 2004-2009 Martin Jericho
// http://jericho.htmlparser.net/
//
// This library is free software; you can redistribute it and/or
// modify it under the terms of either one of the following licences:
//
// 1. The Eclipse Public License (EPL) version 1.0,
// included in this distribution in the file licence-epl-1.0.html
// or available at http://www.eclipse.org/legal/epl-v10.html
//
// 2. The GNU Lesser General Public License (LGPL) version 2.1 or later,
// included in this distribution in the file licence-lgpl-2.1.txt
// or available at http://www.gnu.org/licenses/lgpl.txt
//
// This library is distributed on an "AS IS" basis,
// WITHOUT WARRANTY OF ANY KIND, either express or implied.
// See the individual licence texts for more details.

package net.htmlparser.jericho;

import java.util.Iterator;
import java.util.NoSuchElementException;
import java.io.Closeable;
import java.io.Reader;
import java.io.Writer;
import java.io.IOException;
import java.io.InputStream;
import java.io.InputStreamReader;
import java.nio.CharBuffer;
import java.nio.BufferOverflowException;
import java.net.URL;
import java.net.URLConnection;

/**
 * Represents a streamed source HTML document.
 * 

* This class provides a means, via the {@link #iterator()} method, of sequentially parsing every {@linkplain Tag tag}, {@linkplain CharacterReference character reference} * and plain text segment contained within the source document using a minimum amount of memory. *

* In contrast, the standard {@link Source} class stores the entire source text in memory and caches every tag parsed, * resulting in memory problems when attempting to parse very large files. *

* The {@link #iterator() iterator} parses and returns each segment as the source text is streamed in. * Previous segments are discarded for garbage collection. * Source documents up to 2GB in size can be processed, a limit which is imposed by the java language because of its use of the int data type to index string operations. *

* There is however a significant trade-off in functionality when using the StreamedSource class as opposed to the {@link Source} class. * The {@link Tag#getElement()} method is not supported on tags that are returned by the iterator, nor are any methods that use the {@link Element} class in any way. * The {@link Segment#getSource()} method is also not supported. *

* Most of the methods and constructors in this class mirror similarly named methods in the {@link Source} class where the same functionality is available. *

* See the description of the {@link #iterator()} method for a typical usage example of this class. *

* In contrast to a {@link Source} object, the Reader or InputStream specified in the constructor or created implicitly by the constructor * remains open for the life of the StreamedSource object. If the stream is created internally, it is automatically {@linkplain #close() closed} * when the end of the stream is reached or the StreamedSource object is {@linkplain #finalize() finalized}. * However a Reader or InputStream that is specified directly in a constructor is never closed automatically, as it can not be assumed * that the application has no further use for it. It is the user's responsibility to ensure it is closed in this case. * Explicitly calling the {@link #close()} method on the StreamedSource object ensures that all resources used by it are closed, regardless of whether * they were created internally or supplied externally. *

* The functionality provided by StreamedSource is similar to a StAX parser, * but with some important benefits: *

    *
  • * The source document does not have to be valid XML. It can be plain HTML, can contain invalid syntax, undefined entities, * incorrectly nested elements, {@linkplain TagType#isServerTag() server tags}, or anything else that is commonly found in * "tag soup". *
  • * Every single syntactical construct in the source document's original text is included in the iterator, including the * {@linkplain StartTagType#XML_DECLARATION XML declaration}, {@linkplain CharacterReference character references}, {@linkplain StartTagType#COMMENT comments}, * {@linkplain StartTagType#CDATA_SECTION CDATA sections} and {@linkplain TagType#isServerTag() server tags}, * each providing the segment's {@linkplain Segment#getBegin() begin} and {@linkplain Segment#getEnd() end} position in the source document. * This allows an exact copy of the original document to be generated, allowing modifications to be made only where they are explicitly required. * This is not possible with either SAX or * StAX, which to some extent provide interpretations of the content of the XML * instead of the syntactial structures used in the original source document. *
*

* The following table summarises the differences between the StreamedSource, StAX and SAX interfaces. * Note that some of the available features are documented as optional and may not be supported by all implementations of StAX and SAX. *

* *

* * * * * * * * * * * * * * * * * * *
FeatureStreamedSourceStAXSAX
Parse XML
Parse entities without DTD
Automatically validate XML
Parse HTML
Tolerant of syntax or nesting errors
Provide begin and end character positions of each event1
Provide source text of each event
Handle {@linkplain TagType#isServerTag() server tag} events
Handle {@linkplain StartTagType#XML_DECLARATION XML declaration} event
Handle {@linkplain StartTagType#COMMENT comment} events
Handle {@linkplain StartTagType#CDATA_SECTION CDATA section} events
Handle {@linkplain StartTagType#DOCTYPE_DECLARATION document type declaration} event
Handle {@linkplain CharacterReference character reference} events
Allow chunking of plain text
Allow chunking of comment text
Allow chunking of CDATA section text
Allow specification of maximum buffer size
* 1 StAX optionally reports the "offset" of each event but this could be either byte or character position depending on the source. *

* Note that the {@link OutputDocument} class can not be used to create a modified version of a streamed source document. * Instead, the output document must be constructed manually from the segments provided by the {@link #iterator() iterator}. *

* StreamedSource objects are not thread safe. */ public final class StreamedSource implements Iterable, Closeable { private final StreamedText streamedText; private final StreamedParseText streamedParseText; private final Source source; private final Closeable closeable; // internally created closeable object should be cleaned up internally. private final boolean automaticClose; private boolean coalescing=false; private boolean handleTags=true; private boolean isInitialised=false; private Segment currentSegment=null; private Segment nextParsedSegment=START_SEGMENT; private boolean isXML; private static final boolean assumeNoNestedTags=false; private static final Segment START_SEGMENT=new Segment(-1,-1); private StreamedSource(final Reader reader, final boolean automaticClose, final String encoding, final String encodingSpecificationInfo, final String preliminaryEncodingInfo) throws IOException { closeable=reader; this.automaticClose=automaticClose; streamedText=new StreamedText(reader); streamedParseText=new StreamedParseText(streamedText); source=new Source(streamedText,streamedParseText,encoding,encodingSpecificationInfo,preliminaryEncodingInfo); } private StreamedSource(final EncodingDetector encodingDetector, final boolean automaticClose) throws IOException { this(encodingDetector.openReader(),automaticClose,encodingDetector.getEncoding(),encodingDetector.getEncodingSpecificationInfo(),encodingDetector.getPreliminaryEncoding()+": "+encodingDetector.getPreliminaryEncodingSpecificationInfo()); } /** * Constructs a new StreamedSource object by loading the content from the specified Reader. *

* If the specified reader is an instance of InputStreamReader, the {@link #getEncoding()} method of the * created StreamedSource object returns the encoding from InputStreamReader.getEncoding(). * * @param reader the java.io.Reader from which to load the source text. * @throws java.io.IOException if an I/O error occurs. */ public StreamedSource(final Reader reader) throws IOException { this(reader,false,(reader instanceof InputStreamReader) ? ((InputStreamReader)reader).getEncoding() : null,(reader instanceof InputStreamReader) ? "InputStreamReader.getEncoding() of constructor argument" : null,null); } /** * Constructs a new StreamedSource object by loading the content from the specified InputStream. *

* The algorithm for detecting the character {@linkplain #getEncoding() encoding} of the source document from the raw bytes * of the specified input stream is the same as that for the {@link Source#Source(URLConnection) Source(URLConnection)} constructor of the {@link Source} class, * except that the first step is not possible as there is no * Content-Type header to check. *

* If the specified InputStream does not support the mark method, the algorithm that determines the encoding may have to wrap it * in a BufferedInputStream in order to look ahead at the encoding meta data. * This extra layer of buffering will then remain in place for the life of the StreamedSource, possibly impacting memory usage and/or degrading performance. * It is always preferable to use the {@link #StreamedSource(Reader)} constructor if the encoding is known in advance. * * @param inputStream the java.io.InputStream from which to load the source text. * @throws java.io.IOException if an I/O error occurs. * @see #getEncoding() */ public StreamedSource(final InputStream inputStream) throws IOException { this(new EncodingDetector(inputStream),false); } /** * Constructs a new StreamedSource object by loading the content from the specified URL. *

* This is equivalent to {@link #StreamedSource(URLConnection) StreamedSource(url.openConnection())}. * * @param url the URL from which to load the source text. * @throws java.io.IOException if an I/O error occurs. * @see #getEncoding() */ public StreamedSource(final URL url) throws IOException { this(new EncodingDetector(url.openConnection()),true); } /** * Constructs a new StreamedSource object by loading the content from the specified URLConnection. *

* The algorithm for detecting the character {@linkplain #getEncoding() encoding} of the source document is identical to that described in the * {@link Source#Source(URLConnection) Source(URLConnection)} constructor of the {@link Source} class. *

* The algorithm that determines the encoding may have to wrap the input stream in a BufferedInputStream in order to look ahead * at the encoding meta data if the encoding is not specified in the HTTP headers. * This extra layer of buffering will then remain in place for the life of the StreamedSource, possibly impacting memory usage and/or degrading performance. * It is always preferable to use the {@link #StreamedSource(Reader)} constructor if the encoding is known in advance. * * @param urlConnection the URL connection from which to load the source text. * @throws java.io.IOException if an I/O error occurs. * @see #getEncoding() */ public StreamedSource(final URLConnection urlConnection) throws IOException { this(new EncodingDetector(urlConnection),true); } /** * Constructs a new StreamedSource object from the specified text. *

* Although the CharSequence argument of this constructor apparently contradicts the notion of streaming in the source text, * it can still benefits over the equivalent use of the standard {@link Source} class. *

* Firstly, using the StreamedSource class to iterate the nodes of an in-memory CharSequence source document still requires much less memory * than the equivalent operation using the standard {@link Source} class. *

* Secondly, the specified CharSequence object could possibly implement its own paging mechanism to minimise memory usage. *

* If the specified CharSequence is mutable, its state must not be modified while the StreamedSource is in use. * * @param text the source text. */ public StreamedSource(final CharSequence text) { closeable=null; automaticClose=false; streamedText=new StreamedText(text); streamedParseText=new StreamedParseText(streamedText); source=new Source(text,streamedParseText,null,"Document specified encoding can not be determined automatically from a streamed source",null); } /** * Specifies an existing character array to use for buffering the incoming character stream. *

* The specified buffer is fixed for the life of the StreamedSource object, in contrast to the default buffer which can be automatically replaced * by a larger buffer as needed. * This means that if a {@linkplain Tag tag} (including a {@linkplain StartTagType#COMMENT comment} or {@linkplain StartTagType#CDATA_SECTION CDATA section}) is * encountered that is larger than the specified buffer, an unrecoverable BufferOverflowException is thrown. * This exception is also thrown if {@link #setCoalescing(boolean) coalescing} has been enabled and a plain text segment is encountered * that is larger than the specified buffer. *

* In general this method should only be used if there needs to be an absolute maximum memory limit imposed on the parser, where that requirement is more important * than the ability to parse any source document successfully. *

* This method can only be called before the {@link #iterator()} method has been called. * * @param buffer an existing character array to use for buffering the incoming character stream, must not be null. * @return this StreamedSource instance, allowing multiple property setting methods to be chained in a single statement. * @throws IllegalStateException if the {@link #iterator()} method has already been called. */ public StreamedSource setBuffer(char[] buffer) { if (isInitialised) throw new IllegalStateException("setBuffer() can only be called before iterator() is called"); streamedText.setBuffer(buffer); return this; } /** * Specifies whether an unbroken section of plain text in the source document should always be coalesced into a single {@link Segment} by the {@linkplain #iterator() iterator}. *

* If this property is set to the default value of false, * and a section of plain text is encountered in the document that is larger than the current {@linkplain #getBufferSize() buffer size}, * the text is chunked into multiple consecutive plain text segments in order to minimise memory usage. *

* If this property is set to true then chunking is disabled, ensuring that consecutive plain text segments are never generated, * but instead forcing the internal buffer to expand to fit the largest section of plain text. *

* Note that {@link CharacterReference} segments are always handled separately from plain text, regardless of the value of this property. * For this reason, algorithms that process element content almost always have to be designed to expect the text in multiple segments * in order to handle character references, so there is usually no advantage in {@linkplain #setCoalescing(boolean) coalescing} plain text segments. * * @param coalescing the new value of the coalescing property. * @return this StreamedSource instance, allowing multiple property setting methods to be chained in a single statement. * @throws IllegalStateException if the {@link #iterator()} method has already been called. */ public StreamedSource setCoalescing(final boolean coalescing) { if (isInitialised) throw new IllegalStateException("setPlainTextWriter() can only be called before iterator() is called"); this.coalescing=coalescing; return this; } /** * Closes the underlying Reader or InputStream and releases any system resources associated with it. *

* If the stream is already closed then invoking this method has no effect. * * @throws IOException if an I/O error occurs. */ public void close() throws IOException { if (closeable!=null) closeable.close(); } /** * Returns the character encoding scheme of the source byte stream used to create this object. *

* This method works in essentially the same way as the {@link Source#getEncoding()} method. *

* If the byte stream used to create this object does not support the mark method, the algorithm that determines the encoding may have to wrap it * in a BufferedInputStream in order to look ahead at the encoding meta data. * This extra layer of buffering will then remain in place for the life of the StreamedSource, possibly impacting memory usage and/or degrading performance. * It is always preferable to use the {@link #StreamedSource(Reader)} constructor if the encoding is known in advance. *

* The {@link #getEncodingSpecificationInfo()} method returns a simple description of how the value of this method was determined. * * @return the character encoding scheme of the source byte stream used to create this object, or null if the encoding is not known. * @see #getEncodingSpecificationInfo() */ public String getEncoding() { return source.getEncoding(); } /** * Returns a concise description of how the {@linkplain #getEncoding() encoding} of the source document was determined. *

* The description is intended for informational purposes only. * It is not guaranteed to have any particular format and can not be reliably parsed. * * @return a concise description of how the {@linkplain #getEncoding() encoding} of the source document was determined. * @see #getEncoding() */ public String getEncodingSpecificationInfo() { return source.getEncodingSpecificationInfo(); } /** * Returns the preliminary encoding of the source document together with a concise description of how it was determined. *

* This method works in essentially the same way as the {@link Source#getPreliminaryEncodingInfo()} method. *

* The description returned by this method is intended for informational purposes only. * It is not guaranteed to have any particular format and can not be reliably parsed. * * @return the preliminary encoding of the source document together with a concise description of how it was determined, or null if no preliminary encoding was required. * @see #getEncoding() */ public String getPreliminaryEncodingInfo() { return source.getPreliminaryEncodingInfo(); } /** * Returns an iterator over every {@linkplain Tag tag}, {@linkplain CharacterReference character reference} and plain text segment contained within the source document. *

* Plain text is defined as all text that is not part of a {@link Tag} or {@link CharacterReference}. *

* This results in a sequential walk-through of the entire source document. * The {@linkplain Segment#getEnd() end} position of each segment should correspond with the {@linkplain Segment#getBegin() begin} position of the subsequent segment, * unless any of the tags are enclosed by other tags. * This could happen if there are {@linkplain TagType#isServerTag() server tags} present in the document, or in rare circumstances where the * {@linkplain StartTagType#DOCTYPE_DECLARATION document type declaration} contains {@linkplain StartTagType#MARKUP_DECLARATION markup declarations}. *

* Each segment generated by the iterator is parsed as the source text is streamed in. Previous segments are discarded for garbage collection. *

* If a section of plain text is encountered in the document that is larger than the current {@linkplain #getBufferSize() buffer size}, * the text is chunked into multiple consecutive plain text segments in order to minimise memory usage. * Setting the {@link #setCoalescing(boolean) Coalescing} property to true disables chunking, ensuring that consecutive plain text segments are never generated, * but instead forcing the internal buffer to expand to fit the largest section of plain text. * Note that {@link CharacterReference} segments are always handled separately from plain text, regardless of whether {@linkplain #setCoalescing(boolean) coalescing} * is enabled. For this reason, algorithms that process element content almost always have to be designed to expect the text in multiple segments * in order to handle character references, so there is usually no advantage in {@linkplain #setCoalescing(boolean) coalescing} plain text segments. *

* Character references that are found inside tags, such as those present inside attribute values, do not generate separate segments from the iterator. *

* This method may only be called once on any particular StreamedSource instance. *

*

*
Example:
*
*

* The following code demonstrates the typical (implied) usage of this method through the Iterable interface * to make an exact copy of the document from reader to writer (assuming no server tags are present): *

*
	 * StreamedSource streamedSource=new StreamedSource(reader);
	 * for (Segment segment : streamedSource) {
	 *   if (segment instanceof Tag) {
	 *     Tag tag=(Tag)segment;
	 *     // HANDLE TAG
	 *     // Uncomment the following line to ensure each tag is valid XML:
	 *     // writer.write(tag.tidy()); continue;
	 *   } else if (segment instanceof CharacterReference) {
	 *     CharacterReference characterReference=(CharacterReference)segment;
	 *     // HANDLE CHARACTER REFERENCE
	 *     // Uncomment the following line to decode all character references instead of copying them verbatim:
	 *     // characterReference.appendCharTo(writer); continue;
	 *   } else {
	 *     // HANDLE PLAIN TEXT
	 *   }
	 *   // unless specific handling has prevented getting to here, simply output the segment as is:
	 *   writer.write(segment.toString());
	 * }
*

Note that the last line writer.write(segment.toString()) in the above code can be replaced with the following for improved performance:

*
	 * CharBuffer charBuffer=streamedSource.getCurrentSegmentCharBuffer();
	 * writer.write(charBuffer.array(),charBuffer.position(),charBuffer.length());
*
*
*

* The following code demonstrates how to process the plain text content of a specific element, in this case to print the content of every paragraph element: *

*
	 * StreamedSource streamedSource=new StreamedSource(reader);
	 * StringBuilder sb=new StringBuilder();
	 * boolean insideParagraphElement=false;
	 * for (Segment segment : streamedSource) {
	 *   if (segment instanceof Tag) {
	 *     Tag tag=(Tag)segment;
	 *     if (tag.getName().equals("p")) {
	 *       if (tag instanceof StartTag) {
	 *         insideParagraphElement=true;
	 *         sb.setLength(0);
	 *       } else { // tag instanceof EndTag
	 *         insideParagraphElement=false;
	 *         System.out.println(sb.toString());
	 *       }
	 *     }
	 *   } else if (insideParagraphElement) {
	 *     if (segment instanceof CharacterReference) {
	 *       ((CharacterReference)segment).appendCharTo(sb);
	 *     } else {
	 *       sb.append(segment);
	 *     }
	 *   }
	 * }
*
*
* @return an iterator over every {@linkplain Tag tag}, {@linkplain CharacterReference character reference} and plain text segment contained within the source document. */ public Iterator iterator() { if (isInitialised) throw new IllegalStateException("iterator() can only be called once"); isInitialised=true; return new StreamedSourceIterator(); } /** * Returns the current {@link Segment} from the {@linkplain #iterator()}. *

* This is defined as the last {@link Segment} returned from the iterator's next() method. *

* This method returns null if the iterator's next() method has never been called, or its * hasNext() method has returned the value false. * * @return the current {@link Segment} from the {@linkplain #iterator()}. */ public Segment getCurrentSegment() { return currentSegment; } /** * Returns a CharBuffer containing the source text of the {@linkplain #getCurrentSegment() current segment}. *

* The returned CharBuffer provides a window into the internal char[] buffer including the position and length that spans the * {@linkplain #getCurrentSegment() current segment}. *

* For example, the following code writes the source text of the current segment to writer: *

* CharBuffer charBuffer=streamedSource.getCurrentSegmentCharBuffer();
* writer.write(charBuffer.array(),charBuffer.position(),charBuffer.length()); *

* This may provide a performance benefit over the standard way of accessing the source text of the current segment, * which is to use the CharSequence interface of the segment directly, or to call {@link Segment#toString()}. *

* Because this CharBuffer is a direct window into the internal buffer of the StreamedSource, the contents of the * CharBuffer.array() must not be modified, and the array is only guaranteed to hold the segment source text until the * iterator's hasNext() or next() method is next called. * * @return a CharBuffer containing the source text of the {@linkplain #getCurrentSegment() current segment}. */ public CharBuffer getCurrentSegmentCharBuffer() { return streamedText.getCharBuffer(currentSegment.getBegin(),currentSegment.end); } /** * Indicates whether the source document is likely to be XML. *

* The algorithm used to determine this is designed to be relatively inexpensive and to provide an accurate result in * most normal situations. * An exact determination of whether the source document is XML would require a much more complex analysis of the text. *

* The algorithm is as follows: *

    *
  1. If the document begins with an {@linkplain StartTagType#XML_DECLARATION XML declaration}, it is an XML document. *
  2. If the document begins with a {@linkplain StartTagType#DOCTYPE_DECLARATION document type declaration} that contains the text * "xhtml", it is an XHTML document, and hence * also an XML document. *
  3. If none of the above conditions are met, assume the document is normal HTML, and therefore not an XML document. *
*

* This method can only be called after the {@link #iterator()} method has been called. * * @return true if the source document is likely to be XML, otherwise false. * @throws IllegalStateException if the {@link #iterator()} method has not yet been called. */ public boolean isXML() { if (!isInitialised) throw new IllegalStateException("isXML() method only available after iterator() has been called"); return isXML; } /** * Sets the {@link Logger} that handles log messages. *

* Specifying a null argument disables logging completely for operations performed on this StreamedSource object. *

* A logger instance is created automatically for each StreamedSource object in the same way as is described in the * {@link Source#setLogger(Logger)} method. * * @param logger the logger that will handle log messages, or null to disable logging. * @see Config#LoggerProvider */ public void setLogger(final Logger logger) { source.setLogger(logger); } /** * Returns the {@link Logger} that handles log messages. *

* A logger instance is created automatically for each StreamedSource object using the {@link LoggerProvider} * specified by the static {@link Config#LoggerProvider} property. * This can be overridden by calling the {@link #setLogger(Logger)} method. * The name used for all automatically created logger instances is "net.htmlparser.jericho". * * @return the {@link Logger} that handles log messages, or null if logging is disabled. */ public Logger getLogger() { return source.getLogger(); } /** * Returns the current size of the internal character buffer. *

* This information is generally useful only for investigating memory and performance issues. * * @return the current size of the internal character buffer. */ public int getBufferSize() { return streamedText.getBuffer().length; } /** * Returns a string representation of the object as generated by the default Object.toString() implementation. *

* In contrast to the {@link Source#toString()} implementation, it is generally not possible for this method to return the entire source text. * * @return a string representation of the object as generated by the default Object.toString() implementation. */ public String toString() { return super.toString(); } /** * Called by the garbage collector on an object when garbage collection determines that there are no more references to the object. *

* This implementation calls the {@link #close()} method if the underlying Reader or InputStream stream was created internally. */ protected void finalize() { automaticClose(); } StreamedSource setHandleTags(final boolean handleTags) { this.handleTags=handleTags; return this; } StreamedSource setSearchBegin(final int begin) { if (isInitialised) throw new IllegalStateException("setSearchBegin() can only be called before iterator() is called"); final int segmentEnd=begin-1; nextParsedSegment=new Segment(segmentEnd,segmentEnd); return this; } private void automaticClose() { if (automaticClose) try {close();} catch (IOException ex) {} } private static boolean isXML(final Segment firstNonTextSegment) { if (firstNonTextSegment==null || !(firstNonTextSegment instanceof Tag)) return false; Tag tag=(Tag)firstNonTextSegment; if (tag.getTagType()==StartTagType.XML_DECLARATION) return true; // if document has a DOCTYPE declaration and it contains the text "xhtml", it is an XML document: if (tag.source.getParseText().indexOf("xhtml",tag.begin,tag.end)!=-1) return true; return false; } private class StreamedSourceIterator implements Iterator { private final boolean coalescing; private final boolean handleTags; private Segment nextSegment; private int plainTextSegmentBegin=0; private final char[] charByRef=new char[1]; // used to pass a single character by reference public StreamedSourceIterator() { coalescing=StreamedSource.this.coalescing; handleTags=StreamedSource.this.handleTags; loadNextParsedSegment(); isXML=isXML(nextParsedSegment); } public boolean hasNext() { if (nextSegment==Tag.NOT_CACHED) loadNextParsedSegment(); return nextSegment!=null; } public Segment next() { if (!hasNext()) throw new NoSuchElementException(); final Segment result=nextSegment; nextSegment=(result==nextParsedSegment) ? Tag.NOT_CACHED : nextParsedSegment; streamedText.setMinRequiredBufferBegin(result.end); // guaranteed not to be discarded until next call to loadNextParsedSegment() currentSegment=result; return result; } public void remove() { throw new UnsupportedOperationException(); } private final void loadNextParsedSegment() { nextParsedSegment=findNextParsedSegment(); final int plainTextSegmentEnd=(nextParsedSegment!=null) ? nextParsedSegment.begin : streamedText.length(); nextSegment=(plainTextSegmentBegin=source.fullSequentialParseData[0]) { // do not handle character references inside tags or script elements final CharacterReference characterReference=CharacterReference.construct(source,i,Config.UnterminatedCharacterReferenceSettings.ACCEPT_ALL); if (characterReference!=null) return characterReference; } } else if (handleTags && ch=='<') { final Tag tag=TagType.getTagAt(source,i,false,assumeNoNestedTags); if (tag!=null && !tag.isUnregistered()) { final TagType tagType=tag.getTagType(); if (tag.end>source.fullSequentialParseData[0] && tagType!=StartTagType.DOCTYPE_DECLARATION) { source.fullSequentialParseData[0]=(tagType==StartTagType.NORMAL && tag.name==HTMLElementName.SCRIPT) ? Integer.MAX_VALUE : tag.end; } return tag; } } i++; } if (i





© 2015 - 2024 Weber Informatics LLC | Privacy Policy