
org.jwat.tools.tasks.extract.ExtractFile Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of jwat-tools Show documentation
Show all versions of jwat-tools Show documentation
JWAT-Tools uses the available JWAT libraries to make high level tasks available either from command-line or programmatically.
Common tasks include: Test, Compress, Decompress, CDX, Arc2Warc.
More specialised tasks include: Changed, ContainerMD, Delete, Extract, Interval, PathIndex, Unpack, Headers2CDX.
package org.jwat.tools.tasks.extract;
import java.io.File;
import java.io.FileOutputStream;
import java.io.IOException;
import java.io.InputStream;
import org.jwat.arc.ArcRecordBase;
import org.jwat.archive.ArchiveParser;
import org.jwat.archive.ArchiveParserCallback;
import org.jwat.common.HttpHeader;
import org.jwat.common.Payload;
import org.jwat.common.UriProfile;
import org.jwat.gzip.GzipEntry;
import org.jwat.warc.WarcRecord;
public class ExtractFile implements ArchiveParserCallback {
protected ExtractOptions options;
protected File srcFile;
protected String fileName;
protected int recordNr = 1;
protected byte[] tmpBuf = new byte[8192];
protected long consumed = 0;
public ExtractFile() {
}
public void processFile(File file, ExtractOptions options) {
fileName = file.getName();
ArchiveParser archiveParser = new ArchiveParser();
archiveParser.uriProfile = UriProfile.RFC3986_ABS_16BIT_LAX;
archiveParser.bBlockDigestEnabled = true;
archiveParser.bPayloadDigestEnabled = true;
consumed = archiveParser.parse(file, this);
}
@Override
public void apcFileId(File file, int fileId) {
}
@Override
public void apcGzipEntryStart(GzipEntry gzipEntry, long startOffset) {
}
@Override
public void apcArcRecordStart(ArcRecordBase arcRecord, long startOffset, boolean compressed) throws IOException {
if (options.targetUri == null || options.targetUri.equalsIgnoreCase(arcRecord.header.urlStr)) {
Payload payload = arcRecord.getPayload();
HttpHeader httpHeader = null;
InputStream payloadStream = null;
if (payload != null) {
httpHeader = arcRecord.getHttpHeader();
if (httpHeader != null ) {
payloadStream = httpHeader.getPayloadInputStream();
} else {
payloadStream = payload.getInputStreamComplete();
}
}
if (payloadStream != null) {
FileOutputStream out = new FileOutputStream(new File("extracted." + recordNr), false);
int read;
while ((read = payloadStream.read(tmpBuf)) != -1) {
out.write(tmpBuf, 0, read);
}
out.flush();
out.close();
payloadStream.close();
}
if (httpHeader != null) {
httpHeader.close();
}
if (payload != null) {
payload.close();
}
arcRecord.close();
++recordNr;
}
}
@Override
public void apcWarcRecordStart(WarcRecord warcRecord, long startOffset, boolean compressed) throws IOException {
if (options.targetUri == null || (warcRecord.header.warcTargetUriStr != null && options.targetUri.equalsIgnoreCase(warcRecord.header.warcTargetUriStr))) {
Payload payload = warcRecord.getPayload();
HttpHeader httpHeader = null;
InputStream payloadStream = null;
if (payload != null) {
httpHeader = warcRecord.getHttpHeader();
if (httpHeader != null ) {
payloadStream = httpHeader.getPayloadInputStream();
} else {
payloadStream = payload.getInputStreamComplete();
}
}
if (payloadStream != null) {
FileOutputStream out = new FileOutputStream(new File("extracted." + recordNr), false);
int read;
while ((read = payloadStream.read(tmpBuf)) != -1) {
out.write(tmpBuf, 0, read);
}
out.flush();
out.close();
payloadStream.close();
}
if (httpHeader != null) {
httpHeader.close();
}
if (payload != null) {
payload.close();
}
warcRecord.close();
++recordNr;
}
}
@Override
public void apcUpdateConsumed(long consumed) {
}
@Override
public void apcRuntimeError(Throwable t, long offset, long consumed) {
}
@Override
public void apcDone() {
}
}
© 2015 - 2025 Weber Informatics LLC | Privacy Policy