/******************************************************************************* * Copyright (c) 2000, 2009 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.jdt.internal.core.util; import org.eclipse.jdt.internal.compiler.ast.CompilationUnitDeclaration; import org.eclipse.jdt.internal.compiler.classfmt.ClassFileConstants; import org.eclipse.jdt.internal.compiler.impl.CompilerOptions; import org.eclipse.jdt.internal.compiler.parser.Parser; import org.eclipse.jdt.internal.compiler.parser.Scanner; import org.eclipse.jdt.internal.compiler.problem.ProblemReporter; import org.eclipse.jdt.internal.compiler.problem.ProblemSeverities; /** * Internal parser used for parsing source to create DOM AST nodes. * * @since 3.0 */ public class CommentRecorderParser extends Parser { // support for comments int[] commentStops = new int[10]; int[] commentStarts = new int[10]; int commentPtr = -1; // no comment test with commentPtr value -1 protected final static int CommentIncrement = 100; /** * @param problemReporter * @param optimizeStringLiterals */ public CommentRecorderParser(ProblemReporter problemReporter, boolean optimizeStringLiterals) { super(problemReporter, optimizeStringLiterals); } // old javadoc style check which doesn't include all leading comments into declaration // for backward compatibility with 2.1 DOM public void checkComment() { // discard obsolete comments while inside methods or fields initializer (see bug 74369) if (!(this.diet && this.dietInt==0) && this.scanner.commentPtr >= 0) { flushCommentsDefinedPriorTo(this.endStatementPosition); } boolean deprecated = false; boolean checkDeprecated = false; int lastCommentIndex = -1; //since jdk1.2 look only in the last java doc comment... nextComment : for (lastCommentIndex = this.scanner.commentPtr; lastCommentIndex >= 0; lastCommentIndex--){ //look for @deprecated into the first javadoc comment preceeding the declaration int commentSourceStart = this.scanner.commentStarts[lastCommentIndex]; // javadoc only (non javadoc comment have negative start and/or end positions.) if ((commentSourceStart < 0) || (this.modifiersSourceStart != -1 && this.modifiersSourceStart < commentSourceStart) || (this.scanner.commentStops[lastCommentIndex] < 0)) { continue nextComment; } checkDeprecated = true; int commentSourceEnd = this.scanner.commentStops[lastCommentIndex] - 1; //stop is one over // do not report problem before last parsed comment while recovering code... if (this.javadocParser.shouldReportProblems) { this.javadocParser.reportProblems = this.currentElement == null || commentSourceEnd > this.lastJavadocEnd; } else { this.javadocParser.reportProblems = false; } deprecated = this.javadocParser.checkDeprecation(lastCommentIndex); this.javadoc = this.javadocParser.docComment; if (this.currentElement == null) this.lastJavadocEnd = commentSourceEnd; break nextComment; } if (deprecated) { checkAndSetModifiers(ClassFileConstants.AccDeprecated); } // modify the modifier source start to point at the first comment if (lastCommentIndex >= 0 && checkDeprecated) { this.modifiersSourceStart = this.scanner.commentStarts[lastCommentIndex]; if (this.modifiersSourceStart < 0) { this.modifiersSourceStart = -this.modifiersSourceStart; } } } /* (non-Javadoc) * @see org.eclipse.jdt.internal.compiler.parser.Parser#consumeClassHeader() */ protected void consumeClassHeader() { pushOnCommentsStack(0, this.scanner.commentPtr); super.consumeClassHeader(); } /* (non-Javadoc) * @see org.eclipse.jdt.internal.compiler.parser.Parser#consumeEmptyTypeDeclaration() */ protected void consumeEmptyTypeDeclaration() { pushOnCommentsStack(0, this.scanner.commentPtr); super.consumeEmptyTypeDeclaration(); } /* (non-Javadoc) * @see org.eclipse.jdt.internal.compiler.parser.Parser#consumeInterfaceHeader() */ protected void consumeInterfaceHeader() { pushOnCommentsStack(0, this.scanner.commentPtr); super.consumeInterfaceHeader(); } /* (non-Javadoc) * @see org.eclipse.jdt.internal.compiler.parser.Parser#endParse(int) */ protected CompilationUnitDeclaration endParse(int act) { CompilationUnitDeclaration unit = super.endParse(act); if (unit.comments == null) { pushOnCommentsStack(0, this.scanner.commentPtr); unit.comments = getCommentsPositions(); } return unit; } /* (non-Javadoc) * Save all source comments currently stored before flushing them. * @see org.eclipse.jdt.internal.compiler.parser.Parser#flushCommentsDefinedPriorTo(int) */ public int flushCommentsDefinedPriorTo(int position) { int lastCommentIndex = this.scanner.commentPtr; if (lastCommentIndex < 0) return position; // no comment // compute the index of the first obsolete comment int index = lastCommentIndex; int validCount = 0; while (index >= 0){ int commentEnd = this.scanner.commentStops[index]; if (commentEnd < 0) commentEnd = -commentEnd; // negative end position for non-javadoc comments if (commentEnd <= position){ break; } index--; validCount++; } // if the source at <position> is immediately followed by a line comment, then // flush this comment and shift <position> to the comment end. if (validCount > 0){ int immediateCommentEnd = 0; while (index<lastCommentIndex && (immediateCommentEnd = -this.scanner.commentStops[index+1]) > 0){ // only tolerating non-javadoc comments (non-javadoc comment end positions are negative) // is there any line break until the end of the immediate comment ? (thus only tolerating line comment) immediateCommentEnd--; // comment end in one char too far if (org.eclipse.jdt.internal.compiler.util.Util.getLineNumber(position, this.scanner.lineEnds, 0, this.scanner.linePtr) != org.eclipse.jdt.internal.compiler.util.Util.getLineNumber(immediateCommentEnd, this.scanner.lineEnds, 0, this.scanner.linePtr)) break; position = immediateCommentEnd; validCount--; // flush this comment index++; } } if (index < 0) return position; // no obsolete comment pushOnCommentsStack(0, index); // store comment before flushing them switch (validCount) { case 0: // do nothing break; // move valid comment infos, overriding obsolete comment infos case 2: this.scanner.commentStarts[0] = this.scanner.commentStarts[index+1]; this.scanner.commentStops[0] = this.scanner.commentStops[index+1]; this.scanner.commentTagStarts[0] = this.scanner.commentTagStarts[index+1]; this.scanner.commentStarts[1] = this.scanner.commentStarts[index+2]; this.scanner.commentStops[1] = this.scanner.commentStops[index+2]; this.scanner.commentTagStarts[1] = this.scanner.commentTagStarts[index+2]; break; case 1: this.scanner.commentStarts[0] = this.scanner.commentStarts[index+1]; this.scanner.commentStops[0] = this.scanner.commentStops[index+1]; this.scanner.commentTagStarts[0] = this.scanner.commentTagStarts[index+1]; break; default: System.arraycopy(this.scanner.commentStarts, index + 1, this.scanner.commentStarts, 0, validCount); System.arraycopy(this.scanner.commentStops, index + 1, this.scanner.commentStops, 0, validCount); System.arraycopy(this.scanner.commentTagStarts, index + 1, this.scanner.commentTagStarts, 0, validCount); } this.scanner.commentPtr = validCount - 1; return position; } /* * Build a n*2 matrix of comments positions. * For each position, 0 is for start position and 1 for end position of the comment. */ public int[][] getCommentsPositions() { int[][] positions = new int[this.commentPtr+1][2]; for (int i = 0, max = this.commentPtr; i <= max; i++){ positions[i][0] = this.commentStarts[i]; positions[i][1] = this.commentStops[i]; } return positions; } /* (non-Javadoc) * @see org.eclipse.jdt.internal.compiler.parser.Parser#initialize() */ public void initialize(boolean initializeNLS) { super.initialize(initializeNLS); this.commentPtr = -1; } /* (non-Javadoc) * @see org.eclipse.jdt.internal.compiler.parser.Parser#initialize() */ public void initialize() { super.initialize(); this.commentPtr = -1; } /* (non-Javadoc) * Create and store a specific comment recorder scanner. * @see org.eclipse.jdt.internal.compiler.parser.Parser#initializeScanner() */ public void initializeScanner() { this.scanner = new Scanner( false /*comment*/, false /*whitespace*/, this.options.getSeverity(CompilerOptions.NonExternalizedString) != ProblemSeverities.Ignore /*nls*/, this.options.sourceLevel /*sourceLevel*/, this.options.taskTags/*taskTags*/, this.options.taskPriorities/*taskPriorities*/, this.options.isTaskCaseSensitive/*taskCaseSensitive*/); } /* * Push all stored comments in stack. */ private void pushOnCommentsStack(int start, int end) { for (int i=start; i<=end; i++) { // First see if comment hasn't been already stored int scannerStart = this.scanner.commentStarts[i]<0 ? -this.scanner.commentStarts[i] : this.scanner.commentStarts[i]; int commentStart = this.commentPtr == -1 ? -1 : (this.commentStarts[this.commentPtr]<0 ? -this.commentStarts[this.commentPtr] : this.commentStarts[this.commentPtr]); if (commentStart == -1 || scannerStart > commentStart) { int stackLength = this.commentStarts.length; if (++this.commentPtr >= stackLength) { System.arraycopy( this.commentStarts, 0, this.commentStarts = new int[stackLength + CommentIncrement], 0, stackLength); System.arraycopy( this.commentStops, 0, this.commentStops = new int[stackLength + CommentIncrement], 0, stackLength); } this.commentStarts[this.commentPtr] = this.scanner.commentStarts[i]; this.commentStops[this.commentPtr] = this.scanner.commentStops[i]; } } } /* (non-Javadoc) * Save all source comments currently stored before flushing them. * @see org.eclipse.jdt.internal.compiler.parser.Parser#resetModifiers() */ protected void resetModifiers() { pushOnCommentsStack(0, this.scanner.commentPtr); super.resetModifiers(); } }