Copyright (c) 2000, 2013 IBM Corporation and others. This program and the accompanying materials are made available under the terms of the Eclipse Public License 2.0 which accompanies this distribution, and is available at https://www.eclipse.org/legal/epl-2.0/ SPDX-License-Identifier: EPL-2.0 Contributors: IBM Corporation - initial API and implementation
/******************************************************************************* * Copyright (c) 2000, 2013 IBM Corporation and others. * * This program and the accompanying materials * are made available under the terms of the Eclipse Public License 2.0 * which accompanies this distribution, and is available at * https://www.eclipse.org/legal/epl-2.0/ * * SPDX-License-Identifier: EPL-2.0 * * Contributors: * IBM Corporation - initial API and implementation *******************************************************************************/
package org.eclipse.jdt.internal.compiler.util; import java.io.IOException; import java.io.InputStream; import java.util.ArrayList; import java.util.List; @SuppressWarnings({"rawtypes", "unchecked"}) public class ManifestAnalyzer { private static final int START = 0, IN_CLASSPATH_HEADER = 1, // multistate PAST_CLASSPATH_HEADER = 2, SKIPPING_WHITESPACE = 3, READING_JAR = 4, CONTINUING = 5, SKIP_LINE = 6; private static final char[] CLASSPATH_HEADER_TOKEN = "Class-Path:".toCharArray(); //$NON-NLS-1$ private int classpathSectionsCount; private ArrayList calledFilesNames;
Analyzes the manifest contents. The given input stream is read using a UTF-8 encoded reader. If the contents of the input stream is not encoded using a UTF-8 encoding, the analysis will fail.
Params:
  • inputStream – the given input stream.
Throws:
  • IOException – if an exception occurs while analyzing the file
Returns:true if the analysis is successful, false otherwise.
/** * Analyzes the manifest contents. The given input stream is read using a UTF-8 encoded reader. * If the contents of the input stream is not encoded using a UTF-8 encoding, the analysis will fail. * * @param inputStream the given input stream. * * @return <code>true</code> if the analysis is successful, <code>false</code> otherwise. * @throws IOException if an exception occurs while analyzing the file */
public boolean analyzeManifestContents(InputStream inputStream) throws IOException { char[] chars = Util.getInputStreamAsCharArray(inputStream, -1, Util.UTF_8); return analyzeManifestContents(chars); }
Analyzes the manifest contents.
Params:
  • chars – the content of the manifest
Returns:true if the analysis is successful, false otherwise.
/** * Analyzes the manifest contents. * * @param chars the content of the manifest * * @return <code>true</code> if the analysis is successful, <code>false</code> otherwise. */
public boolean analyzeManifestContents(char[] chars) { int state = START, substate = 0; StringBuffer currentJarToken = new StringBuffer(); int currentChar; this.classpathSectionsCount = 0; this.calledFilesNames = null; for (int i = 0, max = chars.length; i < max;) { currentChar = chars[i++]; if (currentChar == '\r') { // skip \r, will consider \n later (see https://bugs.eclipse.org/bugs/show_bug.cgi?id=251079 ) if (i < max) { currentChar = chars[i++]; } } switch (state) { case START: if (currentChar == CLASSPATH_HEADER_TOKEN[0]) { state = IN_CLASSPATH_HEADER; substate = 1; } else { state = SKIP_LINE; } break; case IN_CLASSPATH_HEADER: if (currentChar == '\n') { state = START; } else if (currentChar != CLASSPATH_HEADER_TOKEN[substate++]) { state = SKIP_LINE; } else if (substate == CLASSPATH_HEADER_TOKEN.length) { state = PAST_CLASSPATH_HEADER; } break; case PAST_CLASSPATH_HEADER: if (currentChar == ' ') { state = SKIPPING_WHITESPACE; this.classpathSectionsCount++; } else { return false; } break; case SKIPPING_WHITESPACE: if (currentChar == '\n') { state = CONTINUING; } else if (currentChar != ' ') { currentJarToken.append((char) currentChar); state = READING_JAR; } else { // >>>>>>>>>>>>>>>>>> Add the latest jar read addCurrentTokenJarWhenNecessary(currentJarToken); } break; case CONTINUING: if (currentChar == '\n') { addCurrentTokenJarWhenNecessary(currentJarToken); state = START; } else if (currentChar == ' ') { state = SKIPPING_WHITESPACE; } else if (currentChar == CLASSPATH_HEADER_TOKEN[0]) { addCurrentTokenJarWhenNecessary(currentJarToken); state = IN_CLASSPATH_HEADER; substate = 1; } else if (this.calledFilesNames == null) { // >>>>>>>>>>>>>>>>>> Add the latest jar read addCurrentTokenJarWhenNecessary(currentJarToken); state = START; } else { // >>>>>>>>>>>>>>>>>> Add the latest jar read addCurrentTokenJarWhenNecessary(currentJarToken); state = SKIP_LINE; } break; case SKIP_LINE: if (currentChar == '\n') { state = START; } break; case READING_JAR: if (currentChar == '\n') { // appends token below state = CONTINUING; // >>>>>>>>>>> Add a break to not add the jar yet as it can continue on the next line break; } else if (currentChar == ' ') { // appends token below state = SKIPPING_WHITESPACE; } else { currentJarToken.append((char) currentChar); break; } addCurrentTokenJarWhenNecessary(currentJarToken); break; } } switch (state) { case START: return true; case IN_CLASSPATH_HEADER: return true; case PAST_CLASSPATH_HEADER: return false; case SKIPPING_WHITESPACE: // >>>>>>>>>>>>>>>>>> Add the latest jar read addCurrentTokenJarWhenNecessary(currentJarToken); return true; case CONTINUING: // >>>>>>>>>>>>>>>>>> Add the latest jar read addCurrentTokenJarWhenNecessary(currentJarToken); return true; case SKIP_LINE: if (this.classpathSectionsCount != 0) { if (this.calledFilesNames == null) { return false; } } return true; case READING_JAR: // >>>>>>>>>>>>>>>>>> Add the latest jar read return false; } return true; } // >>>>>>>>>>>>>>>> Method Extracted from analyzeManifestContents in the READING_JAR Block private boolean addCurrentTokenJarWhenNecessary(StringBuffer currentJarToken) { if (currentJarToken != null && currentJarToken.length() > 0) { if (this.calledFilesNames == null) { this.calledFilesNames = new ArrayList(); } this.calledFilesNames.add(currentJarToken.toString()); currentJarToken.setLength(0); return true; } return false; } // <<<<<<<<<<<<<<<<<<<<<< public int getClasspathSectionsCount() { return this.classpathSectionsCount; } public List getCalledFileNames() { return this.calledFilesNames; } }