
org.fife.ui.rsyntaxtextarea.TokenMakerBase Maven / Gradle / Ivy
/*
* 08/26/2004
*
* TokenMakerBase.java - A base class for token makers.
*
* This library is distributed under a modified BSD license. See the included
* RSyntaxTextArea.License.txt file for details.
*/
package org.fife.ui.rsyntaxtextarea;
import javax.swing.Action;
import javax.swing.text.Segment;
/**
* Base class for token makers.
*
* @author Robert Futrell
* @version 1.0
*/
abstract class TokenMakerBase implements TokenMaker {
/**
* The first token in the returned linked list.
*/
protected Token firstToken;
/**
* Used in the creation of the linked list.
*/
protected Token currentToken;
/**
* Used in the creation of the linked list.
*/
protected Token previousToken;
/**
* The factory that gives us our tokens to use.
*/
private TokenFactory tokenFactory;
/**
* Constructor.
*/
public TokenMakerBase() {
firstToken = currentToken = previousToken = null;
tokenFactory = new DefaultTokenFactory();
}
/**
* {@inheritDoc}
*/
public void addNullToken() {
if (firstToken==null) {
firstToken = tokenFactory.createToken();
currentToken = firstToken;
}
else {
currentToken.setNextToken(tokenFactory.createToken());
previousToken = currentToken;
currentToken = currentToken.getNextToken();
}
}
/**
* Adds the token specified to the current linked list of tokens.
*
* @param segment Segment
to get text from.
* @param start Start offset in segment
of token.
* @param end End offset in segment
of token.
* @param tokenType The token's type.
* @param startOffset The offset in the document at which this token
* occurs.
*/
public void addToken(Segment segment, int start, int end, int tokenType,
int startOffset) {
addToken(segment.array, start,end, tokenType, startOffset);
}
/**
* {@inheritDoc}
*/
public void addToken(char[] array, int start, int end, int tokenType,
int startOffset) {
addToken(array, start, end, tokenType, startOffset, false);
}
/**
* Adds the token specified to the current linked list of tokens.
*
* @param array The character array.
* @param start The starting offset in the array.
* @param end The ending offset in the array.
* @param tokenType The token's type.
* @param startOffset The offset in the document at which this token
* occurs.
* @param hyperlink Whether this token is a hyperlink.
*/
public void addToken(char[] array, int start, int end, int tokenType,
int startOffset, boolean hyperlink) {
if (firstToken==null) {
firstToken = tokenFactory.createToken(array, start, end,
startOffset, tokenType);
currentToken = firstToken; // previous token is still null.
}
else {
currentToken.setNextToken(tokenFactory.createToken(array,
start,end, startOffset, tokenType));
previousToken = currentToken;
currentToken = currentToken.getNextToken();
}
currentToken.setHyperlink(hyperlink);
}
/**
* Returns whether this programming language uses curly braces
* ('{' and '}') to denote code blocks.
*
* The default implementation returns false
; subclasses can
* override this method if necessary.
*
* @return Whether curly braces denote code blocks.
*/
public boolean getCurlyBracesDenoteCodeBlocks() {
return false;
}
/**
* Returns an action to handle "insert break" key presses (i.e. Enter).
* The default implementation returns null
. Subclasses
* can override.
*
* @return The default implementation always returns null
.
*/
public Action getInsertBreakAction() {
return null;
}
/**
* {@inheritDoc}
*/
public int getLastTokenTypeOnLine(Segment text, int initialTokenType) {
// Last parameter doesn't matter if we're not painting.
Token t = getTokenList(text, initialTokenType, 0);
while (t.getNextToken()!=null)
t = t.getNextToken();
return t.type;
}
/**
* {@inheritDoc}
*/
public String[] getLineCommentStartAndEnd() {
return null;
}
/**
* Returns whether tokens of the specified type should have "mark
* occurrences" enabled for the current programming language. The default
* implementation returns true if type is {@link Token#IDENTIFIER}.
* Subclasses can override this method to support other token types, such
* as {@link Token#VARIABLE}.
*
* @param type The token type.
* @return Whether tokens of this type should have "mark occurrences"
* enabled.
*/
public boolean getMarkOccurrencesOfTokenType(int type) {
return type==Token.IDENTIFIER;
}
/**
* The default implementation returns false
always. Languages
* that wish to better support auto-indentation can override this method.
*
* @param token The token the previous line ends with.
* @return Whether the next line should be indented.
*/
public boolean getShouldIndentNextLineAfter(Token token) {
return false;
}
/**
* The default implementation returns false
always.
* Subclasses that are highlighting a markup language should override this
* method to return true
.
*
* @return false
always.
*/
public boolean isMarkupLanguage() {
return false;
}
/**
* Deletes the linked list of tokens so we can begin anew. This should
* never have to be called by the programmer, as it is automatically
* called whenever the user calls
* {@link #getLastTokenTypeOnLine(Segment, int)} or
* {@link #getTokenList(Segment, int, int)}.
*/
protected void resetTokenList() {
firstToken = currentToken = previousToken = null;
tokenFactory.resetAllTokens();
}
}