
org.eclipse.jface.text.DefaultTextDoubleClickStrategy Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of org.eclipse.jface.text Show documentation
Show all versions of org.eclipse.jface.text Show documentation
This is org.eclipse.jface.text jar used by Scout SDK
/*******************************************************************************
* Copyright (c) 2000, 2011 IBM Corporation and others.
* All rights reserved. This program and the accompanying materials
* are made available under the terms of the Eclipse Public License v1.0
* which accompanies this distribution, and is available at
* http://www.eclipse.org/legal/epl-v10.html
*
* Contributors:
* IBM Corporation - initial API and implementation
*******************************************************************************/
package org.eclipse.jface.text;
import java.text.CharacterIterator;
import java.util.Locale;
import com.ibm.icu.text.BreakIterator;
/**
* Standard implementation of
* {@link org.eclipse.jface.text.ITextDoubleClickStrategy}.
*
* Selects words using java.text.BreakIterator
for the default
* locale.
*
* @see java.text.BreakIterator
*/
public class DefaultTextDoubleClickStrategy implements ITextDoubleClickStrategy {
/**
* Implements a character iterator that works directly on
* instances of IDocument
. Used to collaborate with
* the break iterator.
*
* @see IDocument
* @since 2.0
*/
static class DocumentCharacterIterator implements CharacterIterator {
/** Document to iterate over. */
private IDocument fDocument;
/** Start offset of iteration. */
private int fOffset= -1;
/** End offset of iteration. */
private int fEndOffset= -1;
/** Current offset of iteration. */
private int fIndex= -1;
/** Creates a new document iterator. */
public DocumentCharacterIterator() {
}
/**
* Configures this document iterator with the document section to be visited.
*
* @param document the document to be iterated
* @param iteratorRange the range in the document to be iterated
*/
public void setDocument(IDocument document, IRegion iteratorRange) {
fDocument= document;
fOffset= iteratorRange.getOffset();
fEndOffset= fOffset + iteratorRange.getLength();
}
/*
* @see CharacterIterator#first()
*/
public char first() {
fIndex= fOffset;
return current();
}
/*
* @see CharacterIterator#last()
*/
public char last() {
fIndex= fOffset < fEndOffset ? fEndOffset -1 : fEndOffset;
return current();
}
/*
* @see CharacterIterator#current()
*/
public char current() {
if (fOffset <= fIndex && fIndex < fEndOffset) {
try {
return fDocument.getChar(fIndex);
} catch (BadLocationException x) {
}
}
return DONE;
}
/*
* @see CharacterIterator#next()
*/
public char next() {
++fIndex;
int end= getEndIndex();
if (fIndex >= end) {
fIndex= end;
return DONE;
}
return current();
}
/*
* @see CharacterIterator#previous()
*/
public char previous() {
if (fIndex == fOffset)
return DONE;
if (fIndex > fOffset)
-- fIndex;
return current();
}
/*
* @see CharacterIterator#setIndex(int)
*/
public char setIndex(int index) {
fIndex= index;
return current();
}
/*
* @see CharacterIterator#getBeginIndex()
*/
public int getBeginIndex() {
return fOffset;
}
/*
* @see CharacterIterator#getEndIndex()
*/
public int getEndIndex() {
return fEndOffset;
}
/*
* @see CharacterIterator#getIndex()
*/
public int getIndex() {
return fIndex;
}
/*
* @see CharacterIterator#clone()
*/
public Object clone() {
DocumentCharacterIterator i= new DocumentCharacterIterator();
i.fDocument= fDocument;
i.fIndex= fIndex;
i.fOffset= fOffset;
i.fEndOffset= fEndOffset;
return i;
}
}
/**
* The document character iterator used by this strategy.
* @since 2.0
*/
private DocumentCharacterIterator fDocIter= new DocumentCharacterIterator();
/**
* The locale specific word break iterator.
* @since 3.7
*/
private BreakIterator fWordBreakIterator;
/**
* The POSIX word break iterator.
*
* Used to workaround ICU bug not treating '.' as word boundary, see
* http://bugs.icu-project.org/trac/ticket/8371 for details.
*
*
* @since 3.7
*/
private BreakIterator fPOSIXWordBreakIterator;
/**
* Creates a new default text double click strategy.
*/
public DefaultTextDoubleClickStrategy() {
}
/*
* @see org.eclipse.jface.text.ITextDoubleClickStrategy#doubleClicked(org.eclipse.jface.text.ITextViewer)
*/
public void doubleClicked(ITextViewer text) {
int offset= text.getSelectedRange().x;
if (offset < 0)
return;
final IDocument document= text.getDocument();
IRegion region= findExtendedDoubleClickSelection(document, offset);
if (region == null)
region= findWord(document, offset);
if (region != null)
text.setSelectedRange(region.getOffset(), region.getLength());
}
/**
* Tries to find a suitable double click selection for the given offset.
*
* Note: This method must return null
if it simply selects the word at
* the given offset.
*
*
* @param document the document
* @param offset the offset
* @return the selection or null
if none to indicate simple word selection
* @since 3.5
*/
protected IRegion findExtendedDoubleClickSelection(IDocument document, int offset) {
return null;
}
/**
* Tries to find the word at the given offset.
*
* @param document the document
* @param offset the offset
* @return the word or null
if none
* @since 3.5
*/
protected IRegion findWord(IDocument document, int offset) {
return findWord(document, offset, getWordBreakIterator());
}
/**
* Returns the locale specific word break iterator.
*
* @return the locale specific word break iterator
* @since 3.7
*/
private BreakIterator getWordBreakIterator() {
if (fWordBreakIterator == null)
fWordBreakIterator= BreakIterator.getWordInstance();
return fWordBreakIterator;
}
/**
* Returns the POSIX word break iterator.
*
*
* Used to workaround ICU bug not treating '.' as word boundary, see
* http://bugs.icu-project.org/trac/ticket/8371 for details.
*
*
* @return the POSIX word break iterator.
* @since 3.7
*/
private BreakIterator getPOSIXWordBreakIterator() {
if (fPOSIXWordBreakIterator == null)
fPOSIXWordBreakIterator= BreakIterator.getWordInstance(new Locale("en", "US", "POSIX")); //$NON-NLS-1$ //$NON-NLS-2$ //$NON-NLS-3$
return fPOSIXWordBreakIterator;
}
/**
* Tries to find the word at the given offset.
*
* @param document the document
* @param offset the offset
* @param wordBreakIterator the word break iterator
* @return the word or null
if none
* @since 3.7
*/
private IRegion findWord(IDocument document, int offset, BreakIterator wordBreakIterator) {
IRegion line;
try {
line= document.getLineInformationOfOffset(offset);
} catch (BadLocationException e) {
return null;
}
if (offset == line.getOffset() + line.getLength())
return null;
fDocIter.setDocument(document, line);
wordBreakIterator.setText(fDocIter);
int start= wordBreakIterator.preceding(offset);
if (start == BreakIterator.DONE)
start= line.getOffset();
int end= wordBreakIterator.following(offset);
if (end == BreakIterator.DONE)
end= line.getOffset() + line.getLength();
if (wordBreakIterator.isBoundary(offset)) {
if (end - offset > offset - start)
start= offset;
else
end= offset;
}
if (end == start)
return null;
int length= end - start;
try {
// Workaround for ICU bug not treating '.' as word boundary, see http://bugs.icu-project.org/trac/ticket/8371 for details.
if (fPOSIXWordBreakIterator != wordBreakIterator && document.get(start, length).indexOf('.') != -1) {
IRegion wordRegion= findWord(document, offset, getPOSIXWordBreakIterator());
if (wordRegion != null) {
int wordStart= wordRegion.getOffset();
int wordEnd= wordStart + wordRegion.getLength();
// Check that no additional breaks besides '.' are introduced
if ((wordStart == start || wordStart > start && document.getChar(wordStart - 1) == '.') && (wordEnd == end || wordEnd < end && document.getChar(wordEnd) == '.'))
return wordRegion;
}
}
} catch (BadLocationException e) {
// Use previously computed word region
}
return new Region(start, length);
}
}
© 2015 - 2025 Weber Informatics LLC | Privacy Policy