2021-01-02 16:15:27 +01:00
|
|
|
/*******************************************************************************
|
|
|
|
* Copyright (c) 2012 IT Just working.
|
|
|
|
* 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:
|
|
|
|
* IT Just working - initial API and implementation
|
|
|
|
*******************************************************************************/
|
|
|
|
package com.minres.scviewer.database.text;
|
|
|
|
|
|
|
|
import java.io.BufferedReader;
|
|
|
|
import java.io.EOFException;
|
|
|
|
import java.io.File;
|
|
|
|
import java.io.FileInputStream;
|
|
|
|
import java.io.IOException;
|
|
|
|
import java.io.InputStream;
|
|
|
|
import java.io.InputStreamReader;
|
|
|
|
import java.util.ArrayList;
|
|
|
|
import java.util.Arrays;
|
|
|
|
import java.util.Collection;
|
|
|
|
import java.util.HashMap;
|
|
|
|
import java.util.List;
|
|
|
|
import java.util.Map;
|
|
|
|
import java.util.TreeMap;
|
|
|
|
import java.util.regex.Matcher;
|
|
|
|
import java.util.regex.Pattern;
|
|
|
|
import java.util.zip.GZIPInputStream;
|
|
|
|
|
|
|
|
import org.mapdb.DB;
|
|
|
|
import org.mapdb.DBMaker;
|
|
|
|
|
|
|
|
import com.minres.scviewer.database.AssociationType;
|
|
|
|
import com.minres.scviewer.database.DataType;
|
|
|
|
import com.minres.scviewer.database.IWaveform;
|
|
|
|
import com.minres.scviewer.database.IWaveformDb;
|
|
|
|
import com.minres.scviewer.database.IWaveformDbLoader;
|
|
|
|
import com.minres.scviewer.database.InputFormatException;
|
|
|
|
import com.minres.scviewer.database.RelationType;
|
|
|
|
import com.minres.scviewer.database.tx.ITxGenerator;
|
|
|
|
|
|
|
|
public class TextDbLoader implements IWaveformDbLoader{
|
|
|
|
|
|
|
|
private Long maxTime=0L;
|
|
|
|
|
|
|
|
IWaveformDb db;
|
|
|
|
|
|
|
|
List<IWaveform> streams;
|
|
|
|
|
|
|
|
Map<String, RelationType> relationTypes=new HashMap<String, RelationType>();
|
|
|
|
|
|
|
|
DB mapDb;
|
|
|
|
|
|
|
|
public TextDbLoader() {
|
|
|
|
}
|
|
|
|
|
|
|
|
@Override
|
|
|
|
public Long getMaxTime() {
|
|
|
|
return maxTime;
|
|
|
|
}
|
|
|
|
|
|
|
|
@Override
|
|
|
|
public Collection<IWaveform> getAllWaves() {
|
|
|
|
return streams;
|
|
|
|
}
|
|
|
|
|
|
|
|
public Map<Long, ITxGenerator> getGeneratorsById() {
|
|
|
|
TreeMap<Long, ITxGenerator> res = new TreeMap<Long, ITxGenerator>();
|
|
|
|
for(IWaveform stream: streams){ for(ITxGenerator it: ((TxStream)stream).getGenerators()){res.put(it.getId(), it);} }
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
|
|
|
static final byte[] x = "scv_tr_stream".getBytes();
|
|
|
|
|
|
|
|
@Override
|
|
|
|
public boolean load(IWaveformDb db, File file) throws InputFormatException {
|
|
|
|
if(file.isDirectory() || !file.exists()) return false;
|
|
|
|
this.db=db;
|
|
|
|
this.streams = new ArrayList<>();
|
2021-01-02 18:04:48 +01:00
|
|
|
TextDbParser parser = new TextDbParser(this);
|
|
|
|
boolean gzipped = isGzipped(file);
|
2021-01-02 16:15:27 +01:00
|
|
|
try {
|
2021-01-02 18:04:48 +01:00
|
|
|
if(!isTxfile(gzipped?new GZIPInputStream(new FileInputStream(file)):new FileInputStream(file)))
|
|
|
|
return false;
|
|
|
|
} catch(Throwable e) {
|
|
|
|
throw new InputFormatException();
|
|
|
|
}
|
|
|
|
File mapDbFile;
|
|
|
|
try {
|
|
|
|
mapDbFile = File.createTempFile("."+file.getName(), null /*"tmp"*/, null /*file.parentFile*/);
|
|
|
|
} catch (IOException e1) {
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
mapDbFile.delete(); // we just need a file name
|
|
|
|
mapDbFile.deleteOnExit();
|
|
|
|
this.mapDb = DBMaker
|
|
|
|
.fileDB(mapDbFile)
|
|
|
|
.fileMmapEnableIfSupported()
|
|
|
|
.fileMmapPreclearDisable()
|
|
|
|
.cleanerHackEnable()
|
|
|
|
.allocateStartSize(64*1024*1024)
|
|
|
|
.allocateIncrement(64*1024*1024)
|
|
|
|
.make();
|
|
|
|
try {
|
|
|
|
parser.parseInput(gzipped?new GZIPInputStream(new FileInputStream(file)):new FileInputStream(file));
|
2021-01-02 17:02:05 +01:00
|
|
|
} catch(IllegalArgumentException|ArrayIndexOutOfBoundsException e) {
|
|
|
|
} catch(Throwable e) {
|
2021-01-02 16:15:27 +01:00
|
|
|
System.out.println("---->>> Exception "+e.toString()+" caught while loading database");
|
|
|
|
e.printStackTrace();
|
2021-01-02 18:04:48 +01:00
|
|
|
return false;
|
2021-01-02 16:15:27 +01:00
|
|
|
}
|
2021-01-02 18:04:48 +01:00
|
|
|
streams.addAll(parser.streamsById.values());
|
2021-01-02 17:02:05 +01:00
|
|
|
return true;
|
2021-01-02 16:15:27 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
private static boolean isTxfile(InputStream istream) {
|
|
|
|
byte[] buffer = new byte[x.length];
|
|
|
|
try {
|
|
|
|
int readCnt = istream.read(buffer, 0, x.length);
|
|
|
|
istream.close();
|
|
|
|
if(readCnt==x.length){
|
|
|
|
for(int i=0; i<x.length; i++)
|
|
|
|
if(buffer[i]!=x[i]) return false;
|
|
|
|
}
|
|
|
|
return true;
|
|
|
|
} catch (IOException e) {
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
private static boolean isGzipped(File f) {
|
|
|
|
try(InputStream is = new FileInputStream(f)) {
|
|
|
|
byte [] signature = new byte[2];
|
|
|
|
int nread = is.read( signature ); //read the gzip signature
|
|
|
|
return nread == 2 && signature[ 0 ] == (byte) 0x1f && signature[ 1 ] == (byte) 0x8b;
|
|
|
|
} catch (IOException e) {
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-01-02 18:04:48 +01:00
|
|
|
public Collection<RelationType> getAllRelationTypes(){
|
|
|
|
return relationTypes.values();
|
2021-01-02 16:15:27 +01:00
|
|
|
}
|
|
|
|
|
2021-01-02 18:04:48 +01:00
|
|
|
static class TextDbParser {
|
|
|
|
static final Pattern scv_tr_stream = Pattern.compile("^scv_tr_stream\\s+\\(ID (\\d+),\\s+name\\s+\"([^\"]+)\",\\s+kind\\s+\"([^\"]+)\"\\)$");
|
|
|
|
static final Pattern scv_tr_generator = Pattern.compile("^scv_tr_generator\\s+\\(ID\\s+(\\d+),\\s+name\\s+\"([^\"]+)\",\\s+scv_tr_stream\\s+(\\d+),$");
|
|
|
|
static final Pattern begin_attribute = Pattern.compile("^begin_attribute \\(ID (\\d+), name \"([^\"]+)\", type \"([^\"]+)\"\\)$");
|
|
|
|
static final Pattern end_attribute = Pattern.compile("^end_attribute \\(ID (\\d+), name \"([^\"]+)\", type \"([^\"]+)\"\\)$");
|
2021-01-02 16:15:27 +01:00
|
|
|
|
2021-01-02 18:04:48 +01:00
|
|
|
HashMap<Long, TxStream> streamsById = new HashMap<Long, TxStream>();
|
|
|
|
HashMap<Long, TxGenerator> generatorsById = new HashMap<Long, TxGenerator>();
|
|
|
|
HashMap<Long, Tx> transactionsById = new HashMap<Long, Tx>();
|
|
|
|
TxGenerator generator = null;
|
|
|
|
Tx transaction = null;
|
|
|
|
boolean endTransaction=false;
|
|
|
|
final TextDbLoader loader;
|
|
|
|
BufferedReader reader =null;
|
|
|
|
|
|
|
|
public TextDbParser(TextDbLoader loader) {
|
|
|
|
super();
|
|
|
|
this.loader = loader;
|
2021-01-02 16:15:27 +01:00
|
|
|
}
|
|
|
|
|
2021-01-02 18:04:48 +01:00
|
|
|
void parseInput(InputStream inputStream) throws IOException{
|
|
|
|
reader = new BufferedReader(new InputStreamReader(inputStream, "UTF-8"));
|
|
|
|
String curLine = reader.readLine();
|
|
|
|
String nextLine = null;
|
|
|
|
while((nextLine=reader.readLine())!=null && curLine!=null) {
|
|
|
|
curLine=parseLine(curLine, nextLine);
|
2021-01-02 16:15:27 +01:00
|
|
|
}
|
2021-01-02 18:04:48 +01:00
|
|
|
if(curLine!=null)
|
|
|
|
parseLine(curLine, nextLine);
|
|
|
|
}
|
|
|
|
|
|
|
|
private String parseLine(String curLine, String nextLine) throws IOException{
|
|
|
|
String[] tokens = curLine.split("\\s+");
|
|
|
|
if("tx_record_attribute".equals(tokens[0])){//matcher = line =~ /^tx_record_attribute\s+(\d+)\s+"([^"]+)"\s+(\S+)\s*=\s*(.+)$/
|
|
|
|
Long id = Long.parseLong(tokens[1]);
|
|
|
|
String name = tokens[2].substring(1, tokens[2].length());
|
|
|
|
DataType type = DataType.valueOf(tokens[3]);
|
|
|
|
String remaining = tokens.length>5?String.join(" ", Arrays.copyOfRange(tokens, 5, tokens.length-1)):"";
|
|
|
|
transactionsById.get(id).getAttributes().add(new TxAttribute(name, type, AssociationType.RECORD, remaining));
|
|
|
|
} else if("tx_begin".equals(tokens[0])){
|
|
|
|
Long id = Long.parseLong(tokens[1]);
|
|
|
|
TxGenerator gen=generatorsById.get(Long.parseLong(tokens[2]));
|
|
|
|
TxStream stream = (TxStream) gen.getStream();
|
|
|
|
stream.setConcurrency(stream.getConcurrency()+1);
|
|
|
|
transaction = new Tx(id, gen.getStream(), gen, Long.parseLong(tokens[3])*stringToScale(tokens[4]));
|
|
|
|
gen.getTransactions().add(transaction);
|
|
|
|
transactionsById.put(id, transaction);
|
|
|
|
loader.maxTime = loader.maxTime>transaction.getBeginTime()?loader.maxTime:transaction.getBeginTime();
|
|
|
|
if(nextLine!=null && nextLine.charAt(0)=='a') {
|
|
|
|
int idx=0;
|
|
|
|
while(nextLine!=null && nextLine.charAt(0)=='a') {
|
|
|
|
String[] attrTokens=nextLine.split("\\s+");
|
|
|
|
TxAttribute attr = new TxAttribute(gen.getBeginAttrs().get(idx), attrTokens[1]);
|
|
|
|
transaction.getAttributes().add(attr);
|
|
|
|
idx++;
|
|
|
|
nextLine=reader.readLine();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
} else if("tx_end".equals(tokens[0])){//matcher = line =~ /^tx_end\s+(\d+)\s+(\d+)\s+(\d+)\s+([munpf]?s)/
|
|
|
|
Long id = Long.parseLong(tokens[1]);
|
|
|
|
transaction = transactionsById.get(id);
|
|
|
|
assert Integer.parseInt(tokens[2])==transaction.getGenerator().getId();
|
|
|
|
transaction.setEndTime(Long.parseLong(tokens[3])*stringToScale(tokens[4]));
|
|
|
|
loader.maxTime = loader.maxTime>transaction.getEndTime()?loader.maxTime:transaction.getEndTime();
|
2021-01-02 16:15:27 +01:00
|
|
|
TxGenerator gen = (TxGenerator) transaction.getGenerator();
|
2021-01-02 18:04:48 +01:00
|
|
|
TxStream stream = (TxStream) gen.getStream();
|
|
|
|
stream.setConcurrency(stream.getConcurrency()-1);
|
|
|
|
if(nextLine!=null && nextLine.charAt(0)=='a') {
|
|
|
|
int idx=0;
|
|
|
|
while(nextLine!=null && nextLine.charAt(0)=='a') {
|
|
|
|
String[] attrTokens=nextLine.split("\\s+");
|
|
|
|
TxAttribute attr = new TxAttribute(gen.getEndAttrs().get(idx), attrTokens[1]);
|
|
|
|
transaction.getAttributes().add(attr);
|
|
|
|
idx++;
|
|
|
|
nextLine=reader.readLine();
|
|
|
|
}
|
2021-01-02 16:15:27 +01:00
|
|
|
}
|
2021-01-02 18:04:48 +01:00
|
|
|
} else if("tx_relation".equals(tokens[0])){//matcher = line =~ /^tx_relation\s+\"(\S+)\"\s+(\d+)\s+(\d+)$/
|
|
|
|
Tx tr2= transactionsById.get(Long.parseLong(tokens[2]));
|
|
|
|
Tx tr1= transactionsById.get(Long.parseLong(tokens[3]));
|
|
|
|
String relType=tokens[1].substring(1, tokens[1].length()-2);
|
|
|
|
if(!loader.relationTypes.containsKey(relType))
|
|
|
|
loader.relationTypes.put(relType, RelationType.create(relType));
|
|
|
|
TxRelation rel = new TxRelation(loader.relationTypes.get(relType), tr1, tr2);
|
|
|
|
tr1.getOutgoingRelations().add(rel);
|
|
|
|
tr2.getIncomingRelations().add(rel);
|
|
|
|
} else if("scv_tr_stream".equals(tokens[0])){
|
|
|
|
Matcher matcher = scv_tr_stream.matcher(curLine);
|
|
|
|
if (matcher.matches()) {
|
|
|
|
Long id = Long.parseLong(matcher.group(1));
|
|
|
|
TxStream stream = new TxStream(loader, id, matcher.group(2), matcher.group(3));
|
|
|
|
streamsById.put(id, stream);
|
|
|
|
}
|
|
|
|
} else if("scv_tr_generator".equals(tokens[0])){
|
|
|
|
Matcher matcher = scv_tr_generator.matcher(curLine);
|
|
|
|
if ((matcher.matches())) {
|
|
|
|
Long id = Long.parseLong(matcher.group(1));
|
|
|
|
TxStream stream=streamsById.get(Long.parseLong(matcher.group(3)));
|
|
|
|
generator=new TxGenerator(id, stream, matcher.group(2));
|
|
|
|
stream.getGenerators().add(generator);
|
|
|
|
generatorsById.put(id, generator);
|
|
|
|
}
|
|
|
|
} else if("begin_attribute".equals(tokens[0])){
|
|
|
|
Matcher matcher = begin_attribute.matcher(curLine);
|
|
|
|
if ((matcher.matches())) {
|
|
|
|
generator.getBeginAttrs().add(TxAttributeType.getAttrType(matcher.group(2), DataType.valueOf(matcher.group(3)), AssociationType.BEGIN));
|
|
|
|
}
|
|
|
|
} else if("end_attribute".equals(tokens[0])){
|
|
|
|
Matcher matcher = end_attribute.matcher(curLine);
|
|
|
|
if ((matcher.matches())) {
|
|
|
|
generator.getEndAttrs().add(TxAttributeType.getAttrType(matcher.group(2), DataType.valueOf(matcher.group(3)), AssociationType.END));
|
|
|
|
}
|
|
|
|
} else if(")".equals(tokens[0])){
|
|
|
|
generator=null;
|
|
|
|
} else if("a".equals(tokens[0])){//matcher = line =~ /^a\s+(.+)$/
|
|
|
|
System.out.println("Don't know what to do with: '"+curLine+"'");
|
|
|
|
} else
|
|
|
|
System.out.println("Don't know what to do with: '"+curLine+"'");
|
|
|
|
return nextLine;
|
|
|
|
}
|
|
|
|
|
|
|
|
private long stringToScale(String scale){
|
|
|
|
String cmp = scale.trim();
|
|
|
|
if("fs".equals(cmp)) return 1L;
|
|
|
|
if("ps".equals(cmp)) return 1000L;
|
|
|
|
if("ns".equals(cmp)) return 1000000L;
|
|
|
|
if("us".equals(cmp)) return 1000000000L;
|
|
|
|
if("ms".equals(cmp)) return 1000000000000L;
|
|
|
|
if("s".equals(cmp) ) return 1000000000000000L;
|
|
|
|
return 1L;
|
|
|
|
}
|
2021-01-02 16:15:27 +01:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|