tikhomirov@10: /*
tikhomirov@530:  * Copyright (c) 2010-2013 TMate Software Ltd
tikhomirov@74:  *  
tikhomirov@74:  * This program is free software; you can redistribute it and/or modify
tikhomirov@74:  * it under the terms of the GNU General Public License as published by
tikhomirov@74:  * the Free Software Foundation; version 2 of the License.
tikhomirov@74:  *
tikhomirov@74:  * This program is distributed in the hope that it will be useful,
tikhomirov@74:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
tikhomirov@74:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
tikhomirov@74:  * GNU General Public License for more details.
tikhomirov@74:  *
tikhomirov@74:  * For information on how to redistribute this software under
tikhomirov@74:  * the terms of a license other than GNU General Public License
tikhomirov@102:  * contact TMate Software at support@hg4j.com
tikhomirov@0:  */
tikhomirov@77: package org.tmatesoft.hg.internal;
tikhomirov@0: 
tikhomirov@80: import static org.tmatesoft.hg.repo.HgRepository.BAD_REVISION;
tikhomirov@74: import static org.tmatesoft.hg.repo.HgRepository.TIP;
tikhomirov@530: import static org.tmatesoft.hg.internal.Internals.REVLOGV1_RECORD_SIZE;
tikhomirov@5: 
tikhomirov@3: import java.io.File;
tikhomirov@2: import java.io.IOException;
tikhomirov@263: import java.util.zip.Inflater;
tikhomirov@0: 
tikhomirov@74: import org.tmatesoft.hg.core.Nodeid;
tikhomirov@300: import org.tmatesoft.hg.repo.HgInternals;
tikhomirov@423: import org.tmatesoft.hg.repo.HgInvalidControlFileException;
tikhomirov@423: import org.tmatesoft.hg.repo.HgInvalidRevisionException;
tikhomirov@423: import org.tmatesoft.hg.repo.HgInvalidStateException;
tikhomirov@80: import org.tmatesoft.hg.repo.HgRepository;
tikhomirov@520: import org.tmatesoft.hg.util.Adaptable;
tikhomirov@74: 
tikhomirov@10: 
tikhomirov@0: /**
tikhomirov@0:  * ? Single RevlogStream per file per repository with accessor to record access session (e.g. with back/forward operations), 
tikhomirov@198:  * or numerous RevlogStream with separate representation of the underlying data (cached, lazy ChunkStream)?
tikhomirov@74:  * 
tikhomirov@0:  * @see http://mercurial.selenic.com/wiki/Revlog
tikhomirov@0:  * @see http://mercurial.selenic.com/wiki/RevlogNG
tikhomirov@74:  *
tikhomirov@74:  * @author Artem Tikhomirov
tikhomirov@74:  * @author TMate Software Ltd.
tikhomirov@0:  */
tikhomirov@0: public class RevlogStream {
tikhomirov@2: 
tikhomirov@198: 	/*
tikhomirov@198: 	 * makes sense for index with inline data only - actual offset of the record in the .i file (record entry + revision * record size))
tikhomirov@198: 	 * 
tikhomirov@198: 	 * long[] in fact (there are 8-bytes field in the revlog)
tikhomirov@198: 	 * However, (a) DataAccess currently doesn't operate with long seek/length
tikhomirov@198: 	 * and, of greater significance, (b) files with inlined data are designated for smaller files,  
tikhomirov@198: 	 * guess, about 130 Kb, and offset there won't ever break int capacity
tikhomirov@198: 	 */
tikhomirov@198: 	private int[] indexRecordOffset;  
tikhomirov@198: 	private int[] baseRevisions;
tikhomirov@2: 	private boolean inline = false;
tikhomirov@3: 	private final File indexFile;
tikhomirov@396: 	private File dataFile;
tikhomirov@10: 	private final DataAccessProvider dataAccess;
tikhomirov@3: 
tikhomirov@10: 	// if we need anything else from HgRepo, might replace DAP parameter with HgRepo and query it for DAP.
tikhomirov@77: 	public RevlogStream(DataAccessProvider dap, File indexFile) {
tikhomirov@10: 		this.dataAccess = dap;
tikhomirov@3: 		this.indexFile = indexFile;
tikhomirov@3: 	}
tikhomirov@2: 
tikhomirov@9: 	/*package*/ DataAccess getIndexStream() {
tikhomirov@534: 		// FIXME post 1.0 must supply a hint that I'll need really few bytes of data (perhaps, at some offset) 
tikhomirov@280: 		// to avoid mmap files when only few bytes are to be read (i.e. #dataLength())
tikhomirov@534: 		return dataAccess.createReader(indexFile);
tikhomirov@0: 	}
tikhomirov@0: 
tikhomirov@9: 	/*package*/ DataAccess getDataStream() {
tikhomirov@534: 		return dataAccess.createReader(getDataFile());
tikhomirov@534: 	}
tikhomirov@534: 	
tikhomirov@534: 	/*package*/ DataSerializer getIndexStreamWriter() {
tikhomirov@534: 		return dataAccess.createWriter(indexFile, true);
tikhomirov@534: 	}
tikhomirov@534: 	
tikhomirov@534: 	/*package*/ DataSerializer getDataStreamWriter() {
tikhomirov@534: 		return dataAccess.createWriter(getDataFile(), true);
tikhomirov@9: 	}
tikhomirov@9: 	
tikhomirov@396: 	/**
tikhomirov@396: 	 * Constructs file object that corresponds to .d revlog counterpart. 
tikhomirov@396: 	 * Note, it's caller responsibility to ensure this file makes any sense (i.e. check {@link #inline} attribute)
tikhomirov@396: 	 */
tikhomirov@396: 	private File getDataFile() {
tikhomirov@396: 		if (dataFile == null) {
tikhomirov@396: 			final String indexName = indexFile.getName();
tikhomirov@396: 			dataFile = new File(indexFile.getParentFile(), indexName.substring(0, indexName.length() - 1) + "d");
tikhomirov@396: 		}
tikhomirov@396: 		return dataFile;
tikhomirov@396: 	}
tikhomirov@396: 	
tikhomirov@396: 	// initialize exception with the file where revlog structure information comes from
tikhomirov@396: 	public HgInvalidControlFileException initWithIndexFile(HgInvalidControlFileException ex) {
tikhomirov@396: 		return ex.setFile(indexFile);
tikhomirov@396: 	}
tikhomirov@396: 
tikhomirov@396: 	// initialize exception with the file where revlog data comes from
tikhomirov@396: 	public HgInvalidControlFileException initWithDataFile(HgInvalidControlFileException ex) {
tikhomirov@396: 		// exceptions are usually raised after read attepmt, hence inline shall be initialized
tikhomirov@396: 		// although honest approach is to call #initOutline() first
tikhomirov@396: 		return ex.setFile(inline ? indexFile : getDataFile());
tikhomirov@396: 	}
tikhomirov@534: 	
tikhomirov@534: 	/*package-private*/String getDataFileName() {
tikhomirov@534: 		// XXX a temporary solution to provide more info to fill in exceptions other than 
tikhomirov@534: 		// HgInvalidControlFileException (those benefit from initWith* methods above)
tikhomirov@534: 		//
tikhomirov@534: 		// Besides, since RevlogStream represents both revlogs with user data (those with WC representative and 
tikhomirov@534: 		// system data under store/data) and system-only revlogs (like changelog and manifest), there's no
tikhomirov@534: 		// easy way to supply human-friendly name of the active file (independent from whether it's index of data)
tikhomirov@534: 		return inline ? indexFile.getPath() : getDataFile().getPath();
tikhomirov@534: 	}
tikhomirov@396: 
tikhomirov@534: 	public boolean isInlineData() {
tikhomirov@534: 		initOutline();
tikhomirov@534: 		return inline;
tikhomirov@534: 	}
tikhomirov@396: 	
tikhomirov@2: 	public int revisionCount() {
tikhomirov@2: 		initOutline();
tikhomirov@198: 		return baseRevisions.length;
tikhomirov@2: 	}
tikhomirov@22: 	
tikhomirov@295: 	/**
tikhomirov@354: 	 * @throws HgInvalidControlFileException if attempt to read index file failed
tikhomirov@354: 	 * @throws HgInvalidRevisionException if revisionIndex argument doesn't represent a valid record in the revlog
tikhomirov@295: 	 */
tikhomirov@354: 	public int dataLength(int revisionIndex) throws HgInvalidControlFileException, HgInvalidRevisionException {
tikhomirov@22: 		// XXX in fact, use of iterate() instead of this implementation may be quite reasonable.
tikhomirov@22: 		//
tikhomirov@354: 		revisionIndex = checkRevisionIndex(revisionIndex);
tikhomirov@280: 		DataAccess daIndex = getIndexStream();
tikhomirov@22: 		try {
tikhomirov@354: 			int recordOffset = getIndexOffsetInt(revisionIndex);
tikhomirov@22: 			daIndex.seek(recordOffset + 12); // 6+2+4
tikhomirov@22: 			int actualLen = daIndex.readInt();
tikhomirov@22: 			return actualLen; 
tikhomirov@22: 		} catch (IOException ex) {
tikhomirov@440: 			throw new HgInvalidControlFileException(null, ex, indexFile).setRevisionIndex(revisionIndex);
tikhomirov@22: 		} finally {
tikhomirov@22: 			daIndex.done();
tikhomirov@22: 		}
tikhomirov@22: 	}
tikhomirov@22: 	
tikhomirov@295: 	/**
tikhomirov@354: 	 * Read nodeid at given index
tikhomirov@354: 	 * 
tikhomirov@354: 	 * @throws HgInvalidControlFileException if attempt to read index file failed
tikhomirov@354: 	 * @throws HgInvalidRevisionException if revisionIndex argument doesn't represent a valid record in the revlog
tikhomirov@295: 	 */
tikhomirov@354: 	public byte[] nodeid(int revisionIndex) throws HgInvalidControlFileException, HgInvalidRevisionException {
tikhomirov@354: 		revisionIndex = checkRevisionIndex(revisionIndex);
tikhomirov@80: 		DataAccess daIndex = getIndexStream();
tikhomirov@80: 		try {
tikhomirov@354: 			int recordOffset = getIndexOffsetInt(revisionIndex);
tikhomirov@80: 			daIndex.seek(recordOffset + 32);
tikhomirov@80: 			byte[] rv = new byte[20];
tikhomirov@80: 			daIndex.readBytes(rv, 0, 20);
tikhomirov@80: 			return rv;
tikhomirov@80: 		} catch (IOException ex) {
tikhomirov@440: 			throw new HgInvalidControlFileException("Revision lookup failed", ex, indexFile).setRevisionIndex(revisionIndex);
tikhomirov@80: 		} finally {
tikhomirov@88: 			daIndex.done();
tikhomirov@88: 		}
tikhomirov@88: 	}
tikhomirov@295: 
tikhomirov@295: 	/**
tikhomirov@295: 	 * Get link field from the index record.
tikhomirov@354: 	 * 
tikhomirov@354: 	 * @throws HgInvalidControlFileException if attempt to read index file failed
tikhomirov@354: 	 * @throws HgInvalidRevisionException if revisionIndex argument doesn't represent a valid record in the revlog
tikhomirov@295: 	 */
tikhomirov@354: 	public int linkRevision(int revisionIndex) throws HgInvalidControlFileException, HgInvalidRevisionException {
tikhomirov@354: 		revisionIndex = checkRevisionIndex(revisionIndex);
tikhomirov@88: 		DataAccess daIndex = getIndexStream();
tikhomirov@88: 		try {
tikhomirov@354: 			int recordOffset = getIndexOffsetInt(revisionIndex);
tikhomirov@88: 			daIndex.seek(recordOffset + 20);
tikhomirov@88: 			int linkRev = daIndex.readInt();
tikhomirov@88: 			return linkRev;
tikhomirov@88: 		} catch (IOException ex) {
tikhomirov@440: 			throw new HgInvalidControlFileException("Linked revision lookup failed", ex, indexFile).setRevisionIndex(revisionIndex);
tikhomirov@88: 		} finally {
tikhomirov@88: 			daIndex.done();
tikhomirov@80: 		}
tikhomirov@80: 	}
tikhomirov@80: 	
tikhomirov@49: 	// Perhaps, RevlogStream should be limited to use of plain int revisions for access,
tikhomirov@49: 	// while Nodeids should be kept on the level up, in Revlog. Guess, Revlog better keep
tikhomirov@49: 	// map of nodeids, and once this comes true, we may get rid of this method.
tikhomirov@80: 	// Unlike its counterpart, {@link Revlog#getLocalRevisionNumber()}, doesn't fail with exception if node not found,
tikhomirov@80: 	/**
tikhomirov@80: 	 * @return integer in [0..revisionCount()) or {@link HgRepository#BAD_REVISION} if not found
tikhomirov@354: 	 * @throws HgInvalidControlFileException if attempt to read index file failed
tikhomirov@80: 	 */
tikhomirov@367: 	public int findRevisionIndex(Nodeid nodeid) throws HgInvalidControlFileException {
tikhomirov@22: 		// XXX this one may be implemented with iterate() once there's mechanism to stop iterations
tikhomirov@22: 		final int indexSize = revisionCount();
tikhomirov@22: 		DataAccess daIndex = getIndexStream();
tikhomirov@22: 		try {
tikhomirov@24: 			byte[] nodeidBuf = new byte[20];
tikhomirov@22: 			for (int i = 0; i < indexSize; i++) {
tikhomirov@22: 				daIndex.skip(8);
tikhomirov@22: 				int compressedLen = daIndex.readInt();
tikhomirov@22: 				daIndex.skip(20);
tikhomirov@24: 				daIndex.readBytes(nodeidBuf, 0, 20);
tikhomirov@24: 				if (nodeid.equalsTo(nodeidBuf)) {
tikhomirov@22: 					return i;
tikhomirov@22: 				}
tikhomirov@22: 				daIndex.skip(inline ? 12 + compressedLen : 12);
tikhomirov@22: 			}
tikhomirov@22: 		} catch (IOException ex) {
tikhomirov@440: 			throw new HgInvalidControlFileException("Revision lookup failed", ex, indexFile).setRevision(nodeid);
tikhomirov@22: 		} finally {
tikhomirov@22: 			daIndex.done();
tikhomirov@22: 		}
tikhomirov@80: 		return BAD_REVISION;
tikhomirov@22: 	}
tikhomirov@538: 	
tikhomirov@538: 	public long newEntryOffset() {
tikhomirov@538: 		if (revisionCount() == 0) {
tikhomirov@538: 			return 0;
tikhomirov@538: 		}
tikhomirov@538: 		DataAccess daIndex = getIndexStream();
tikhomirov@538: 		int lastRev = revisionCount() - 1;
tikhomirov@538: 		try {
tikhomirov@538: 			int recordOffset = getIndexOffsetInt(lastRev);
tikhomirov@538: 			daIndex.seek(recordOffset);
tikhomirov@538: 			long value = daIndex.readLong();
tikhomirov@538: 			value = value >>> 16;
tikhomirov@538: 			int compressedLen = daIndex.readInt();
tikhomirov@538: 			return lastRev == 0 ? compressedLen : value + compressedLen;
tikhomirov@538: 		} catch (IOException ex) {
tikhomirov@538: 			throw new HgInvalidControlFileException("Linked revision lookup failed", ex, indexFile).setRevisionIndex(lastRev);
tikhomirov@538: 		} finally {
tikhomirov@538: 			daIndex.done();
tikhomirov@538: 		}
tikhomirov@538: 	}
tikhomirov@538: 
tikhomirov@22: 
tikhomirov@2: 
tikhomirov@3: 	// should be possible to use TIP, ALL, or -1, -2, -n notation of Hg
tikhomirov@3: 	// ? boolean needsNodeid
tikhomirov@423: 	public void iterate(int start, int end, boolean needData, Inspector inspector) throws HgInvalidRevisionException, HgInvalidControlFileException {
tikhomirov@2: 		initOutline();
tikhomirov@198: 		final int indexSize = revisionCount();
tikhomirov@3: 		if (indexSize == 0) {
tikhomirov@3: 			return;
tikhomirov@3: 		}
tikhomirov@5: 		if (end == TIP) {
tikhomirov@3: 			end = indexSize - 1;
tikhomirov@3: 		}
tikhomirov@5: 		if (start == TIP) {
tikhomirov@5: 			start = indexSize - 1;
tikhomirov@5: 		}
tikhomirov@300: 		HgInternals.checkRevlogRange(start, end, indexSize-1);
tikhomirov@2: 		// XXX may cache [start .. end] from index with a single read (pre-read)
tikhomirov@2: 		
tikhomirov@242: 		ReaderN1 r = new ReaderN1(needData, inspector);
tikhomirov@2: 		try {
tikhomirov@242: 			r.start(end - start + 1);
tikhomirov@242: 			r.range(start, end);
tikhomirov@242: 		} catch (IOException ex) {
tikhomirov@366: 			throw new HgInvalidControlFileException(String.format("Failed reading [%d..%d]", start, end), ex, indexFile);
tikhomirov@366: 		} catch (HgInvalidControlFileException ex) {
tikhomirov@366: 			throw ex;
tikhomirov@242: 		} finally {
tikhomirov@242: 			r.finish();
tikhomirov@242: 		}
tikhomirov@242: 	}
tikhomirov@242: 	
tikhomirov@242: 	/**
tikhomirov@242: 	 * Effective alternative to {@link #iterate(int, int, boolean, Inspector) batch read}, when only few selected 
tikhomirov@242: 	 * revisions are of interest.
tikhomirov@242: 	 * @param sortedRevisions revisions to walk, in ascending order.
tikhomirov@242: 	 * @param needData whether inspector needs access to header only
tikhomirov@242: 	 * @param inspector callback to process entries
tikhomirov@242: 	 */
tikhomirov@366: 	public void iterate(int[] sortedRevisions, boolean needData, Inspector inspector) throws HgInvalidRevisionException, HgInvalidControlFileException /*REVISIT - too general exception*/ {
tikhomirov@242: 		final int indexSize = revisionCount();
tikhomirov@242: 		if (indexSize == 0 || sortedRevisions.length == 0) {
tikhomirov@242: 			return;
tikhomirov@242: 		}
tikhomirov@347: 		if (sortedRevisions[0] > indexSize) {
tikhomirov@347: 			throw new HgInvalidRevisionException(String.format("Can't iterate [%d, %d] in range [0..%d]", sortedRevisions[0], sortedRevisions[sortedRevisions.length - 1], indexSize), null, sortedRevisions[0]);
tikhomirov@347: 		}
tikhomirov@347: 		if (sortedRevisions[sortedRevisions.length - 1] > indexSize) {
tikhomirov@347: 			throw new HgInvalidRevisionException(String.format("Can't iterate [%d, %d] in range [0..%d]", sortedRevisions[0], sortedRevisions[sortedRevisions.length - 1], indexSize), null, sortedRevisions[sortedRevisions.length - 1]);
tikhomirov@242: 		}
tikhomirov@242: 
tikhomirov@242: 		ReaderN1 r = new ReaderN1(needData, inspector);
tikhomirov@242: 		try {
tikhomirov@242: 			r.start(sortedRevisions.length);
tikhomirov@242: 			for (int i = 0; i < sortedRevisions.length; ) {
tikhomirov@242: 				int x = i;
tikhomirov@242: 				i++;
tikhomirov@242: 				while (i < sortedRevisions.length) {
tikhomirov@242: 					if (sortedRevisions[i] == sortedRevisions[i-1] + 1) {
tikhomirov@242: 						i++;
tikhomirov@2: 					} else {
tikhomirov@217: 						break;
tikhomirov@217: 					}
tikhomirov@217: 				}
tikhomirov@242: 				// commitRevisions[x..i-1] are sequential
tikhomirov@242: 				if (!r.range(sortedRevisions[x], sortedRevisions[i-1])) {
tikhomirov@242: 					return;
tikhomirov@51: 				}
tikhomirov@2: 			}
tikhomirov@2: 		} catch (IOException ex) {
tikhomirov@366: 			final int c = sortedRevisions.length;
tikhomirov@366: 			throw new HgInvalidControlFileException(String.format("Failed reading %d revisions in [%d; %d]",c, sortedRevisions[0], sortedRevisions[c-1]), ex, indexFile);
tikhomirov@366: 		} catch (HgInvalidControlFileException ex) {
tikhomirov@423: 			// TODO post-1.0 fill HgRuntimeException with appropriate file (either index or data, depending on error source)
tikhomirov@366: 			throw ex;
tikhomirov@3: 		} finally {
tikhomirov@242: 			r.finish();
tikhomirov@2: 		}
tikhomirov@2: 	}
tikhomirov@198: 
tikhomirov@198: 	private int getBaseRevision(int revision) {
tikhomirov@198: 		return baseRevisions[revision];
tikhomirov@198: 	}
tikhomirov@198: 
tikhomirov@198: 	/**
tikhomirov@354: 	 * @param revisionIndex shall be valid index, [0..baseRevisions.length-1]. 
tikhomirov@354: 	 * It's advised to use {@link #checkRevisionIndex(int)} to ensure argument is correct. 
tikhomirov@198: 	 * @return  offset of the revision's record in the index (.i) stream
tikhomirov@198: 	 */
tikhomirov@354: 	private int getIndexOffsetInt(int revisionIndex) {
tikhomirov@354: 		return inline ? indexRecordOffset[revisionIndex] : revisionIndex * REVLOGV1_RECORD_SIZE;
tikhomirov@354: 	}
tikhomirov@354: 	
tikhomirov@354: 	private int checkRevisionIndex(int revisionIndex) throws HgInvalidRevisionException {
tikhomirov@354: 		final int last = revisionCount() - 1;
tikhomirov@354: 		if (revisionIndex == TIP) {
tikhomirov@354: 			revisionIndex = last;
tikhomirov@354: 		}
tikhomirov@354: 		if (revisionIndex < 0 || revisionIndex > last) {
tikhomirov@354: 			throw new HgInvalidRevisionException(revisionIndex).setRevisionIndex(revisionIndex, 0, last);
tikhomirov@354: 		}
tikhomirov@354: 		return revisionIndex;
tikhomirov@198: 	}
tikhomirov@198: 
tikhomirov@425: 	private void initOutline() throws HgInvalidControlFileException {
tikhomirov@198: 		if (baseRevisions != null && baseRevisions.length > 0) {
tikhomirov@2: 			return;
tikhomirov@2: 		}
tikhomirov@9: 		DataAccess da = getIndexStream();
tikhomirov@2: 		try {
tikhomirov@202: 			if (da.isEmpty()) {
tikhomirov@202: 				// do not fail with exception if stream is empty, it's likely intentional
tikhomirov@202: 				baseRevisions = new int[0];
tikhomirov@202: 				return;
tikhomirov@202: 			}
tikhomirov@9: 			int versionField = da.readInt();
tikhomirov@170: 			da.readInt(); // just to skip next 4 bytes of offset + flags
tikhomirov@2: 			final int INLINEDATA = 1 << 16;
tikhomirov@2: 			inline = (versionField & INLINEDATA) != 0;
tikhomirov@288: 			IntVector resBases, resOffsets = null;
tikhomirov@420: 			int entryCountGuess = Internals.ltoi(da.longLength() / REVLOGV1_RECORD_SIZE);
tikhomirov@288: 			if (inline) {
tikhomirov@288: 				entryCountGuess >>>= 2; // pure guess, assume useful data takes 3/4 of total space
tikhomirov@288: 				resOffsets = new IntVector(entryCountGuess, 5000);
tikhomirov@288: 			}
tikhomirov@288: 			resBases = new IntVector(entryCountGuess, 5000);
tikhomirov@288: 			
tikhomirov@2: 			long offset = 0; // first offset is always 0, thus Hg uses it for other purposes
tikhomirov@9: 			while(true) {
tikhomirov@9: 				int compressedLen = da.readInt();
tikhomirov@5: 				// 8+4 = 12 bytes total read here
tikhomirov@49: 				@SuppressWarnings("unused")
tikhomirov@9: 				int actualLen = da.readInt();
tikhomirov@9: 				int baseRevision = da.readInt();
tikhomirov@5: 				// 12 + 8 = 20 bytes read here
tikhomirov@2: //				int linkRevision = di.readInt();
tikhomirov@2: //				int parent1Revision = di.readInt();
tikhomirov@2: //				int parent2Revision = di.readInt();
tikhomirov@2: //				byte[] nodeid = new byte[32];
tikhomirov@198: 				resBases.add(baseRevision);
tikhomirov@2: 				if (inline) {
tikhomirov@420: 					int o = Internals.ltoi(offset);
tikhomirov@198: 					if (o != offset) {
tikhomirov@198: 						// just in case, can't happen, ever, unless HG (or some other bad tool) produces index file 
tikhomirov@198: 						// with inlined data of size greater than 2 Gb.
tikhomirov@423: 						throw new HgInvalidStateException("Data too big, offset didn't fit to sizeof(int)");
tikhomirov@198: 					}
tikhomirov@198: 					resOffsets.add(o + REVLOGV1_RECORD_SIZE * resOffsets.size());
tikhomirov@9: 					da.skip(3*4 + 32 + compressedLen); // Check: 44 (skip) + 20 (read) = 64 (total RevlogNG record size)
tikhomirov@2: 				} else {
tikhomirov@9: 					da.skip(3*4 + 32);
tikhomirov@2: 				}
tikhomirov@10: 				if (da.isEmpty()) {
tikhomirov@9: 					// fine, done then
tikhomirov@288: 					baseRevisions = resBases.toArray(true);
tikhomirov@198: 					if (inline) {
tikhomirov@288: 						indexRecordOffset = resOffsets.toArray(true);
tikhomirov@198: 					}
tikhomirov@9: 					break;
tikhomirov@10: 				} else {
tikhomirov@10: 					// start reading next record
tikhomirov@10: 					long l = da.readLong();
tikhomirov@10: 					offset = l >>> 16;
tikhomirov@9: 				}
tikhomirov@2: 			}
tikhomirov@2: 		} catch (IOException ex) {
tikhomirov@423: 			throw new HgInvalidControlFileException("Failed to analyze revlog index", ex, indexFile);
tikhomirov@9: 		} finally {
tikhomirov@9: 			da.done();
tikhomirov@2: 		}
tikhomirov@3: 	}
tikhomirov@3: 	
tikhomirov@242: 	/**
tikhomirov@242: 	 * operation with single file open/close and multiple diverse reads.
tikhomirov@366: 	 * XXX initOutline might need similar extraction to keep N1 format knowledge  
tikhomirov@242: 	 */
tikhomirov@242: 	class ReaderN1 {
tikhomirov@242: 		private final Inspector inspector;
tikhomirov@242: 		private final boolean needData;
tikhomirov@242: 		private DataAccess daIndex = null, daData = null;
tikhomirov@242: 		private Lifecycle.BasicCallback cb = null;
tikhomirov@520: 		private Lifecycle lifecycleListener = null;
tikhomirov@242: 		private int lastRevisionRead = BAD_REVISION;
tikhomirov@242: 		private DataAccess lastUserData;
tikhomirov@263: 		// next are to track two major bottlenecks - patch application and actual time spent in inspector 
tikhomirov@264: //		private long applyTime, inspectorTime; // TIMING
tikhomirov@263: 
tikhomirov@242: 
tikhomirov@242: 		public ReaderN1(boolean needData, Inspector insp) {
tikhomirov@242: 			assert insp != null;
tikhomirov@242: 			this.needData = needData;
tikhomirov@242: 			inspector = insp;
tikhomirov@242: 		}
tikhomirov@242: 		
tikhomirov@242: 		public void start(int totalWork) {
tikhomirov@242: 			daIndex = getIndexStream();
tikhomirov@242: 			if (needData && !inline) {
tikhomirov@242: 				daData = getDataStream();
tikhomirov@242: 			}
tikhomirov@520: 			lifecycleListener = Adaptable.Factory.getAdapter(inspector, Lifecycle.class, null);
tikhomirov@520: 			if (lifecycleListener != null) {
tikhomirov@242: 				cb = new Lifecycle.BasicCallback();
tikhomirov@520: 				lifecycleListener.start(totalWork, cb, cb);
tikhomirov@242: 			}
tikhomirov@264: //			applyTime = inspectorTime = 0; // TIMING
tikhomirov@242: 		}
tikhomirov@242: 
tikhomirov@520: 		// invoked only once per instance
tikhomirov@242: 		public void finish() {
tikhomirov@242: 			if (lastUserData != null) {
tikhomirov@242: 				lastUserData.done();
tikhomirov@242: 				lastUserData = null;
tikhomirov@242: 			}
tikhomirov@520: 			if (lifecycleListener != null) {
tikhomirov@520: 				lifecycleListener.finish(cb);
tikhomirov@520: 				lifecycleListener = null;
tikhomirov@520: 				cb = null;
tikhomirov@520: 				
tikhomirov@242: 			}
tikhomirov@242: 			daIndex.done();
tikhomirov@242: 			if (daData != null) {
tikhomirov@242: 				daData.done();
tikhomirov@520: 				daData = null;
tikhomirov@242: 			}
tikhomirov@264: //			System.out.printf("applyTime:%d ms, inspectorTime: %d ms\n", applyTime, inspectorTime); // TIMING
tikhomirov@242: 		}
tikhomirov@263: 
tikhomirov@520: 		// may be invoked few times per instance life
tikhomirov@423: 		public boolean range(int start, int end) throws IOException {
tikhomirov@242: 			byte[] nodeidBuf = new byte[20];
tikhomirov@242: 			int i;
tikhomirov@242: 			// it (i.e. replace with i >= start)
tikhomirov@242: 			if (needData && (i = getBaseRevision(start)) < start) {
tikhomirov@242: 				// if lastRevisionRead in [baseRevision(start), start)  can reuse lastUserData
tikhomirov@242: 				// doesn't make sense to reuse if lastRevisionRead == start (too much to change in the cycle below). 
tikhomirov@242: 				if (lastRevisionRead != BAD_REVISION && i <= lastRevisionRead && lastRevisionRead < start) {
tikhomirov@242: 					i = lastRevisionRead + 1; // start with first not-yet-read revision
tikhomirov@242: 				} else {
tikhomirov@242: 					if (lastUserData != null) {
tikhomirov@242: 						lastUserData.done();
tikhomirov@242: 						lastUserData = null;
tikhomirov@242: 					}
tikhomirov@242: 				}
tikhomirov@242: 			} else {
tikhomirov@242: 				// don't need to clean lastUserData as it's always null when !needData
tikhomirov@242: 				i = start;
tikhomirov@242: 			}
tikhomirov@242: 			
tikhomirov@242: 			daIndex.seek(getIndexOffsetInt(i));
tikhomirov@258: 			//
tikhomirov@263: 			// reuse some instances
tikhomirov@329: 			final Patch patch = new Patch();
tikhomirov@263: 			final Inflater inflater = new Inflater();
tikhomirov@263: 			// can share buffer between instances of InflaterDataAccess as I never read any two of them in parallel
tikhomirov@263: 			final byte[] inflaterBuffer = new byte[1024];
tikhomirov@263: 			//
tikhomirov@242: 			
tikhomirov@242: 			for (; i <= end; i++ ) {
tikhomirov@242: 				if (inline && needData) {
tikhomirov@242: 					// inspector reading data (though FilterDataAccess) may have affected index position
tikhomirov@242: 					daIndex.seek(getIndexOffsetInt(i));
tikhomirov@242: 				}
tikhomirov@242: 				long l = daIndex.readLong(); // 0
tikhomirov@242: 				long offset = i == 0 ? 0 : (l >>> 16);
tikhomirov@242: 				@SuppressWarnings("unused")
tikhomirov@420: 				int flags = (int) (l & 0x0FFFF);
tikhomirov@242: 				int compressedLen = daIndex.readInt(); // +8
tikhomirov@242: 				int actualLen = daIndex.readInt(); // +12
tikhomirov@242: 				int baseRevision = daIndex.readInt(); // +16
tikhomirov@242: 				int linkRevision = daIndex.readInt(); // +20
tikhomirov@242: 				int parent1Revision = daIndex.readInt();
tikhomirov@242: 				int parent2Revision = daIndex.readInt();
tikhomirov@242: 				// Hg has 32 bytes here, uses 20 for nodeid, and keeps 12 last bytes empty
tikhomirov@242: 				daIndex.readBytes(nodeidBuf, 0, 20); // +32
tikhomirov@242: 				daIndex.skip(12);
tikhomirov@242: 				DataAccess userDataAccess = null;
tikhomirov@242: 				if (needData) {
tikhomirov@420: 					long streamOffset;
tikhomirov@242: 					DataAccess streamDataAccess;
tikhomirov@242: 					if (inline) {
tikhomirov@242: 						streamDataAccess = daIndex;
tikhomirov@242: 						streamOffset = getIndexOffsetInt(i) + REVLOGV1_RECORD_SIZE; // don't need to do seek as it's actual position in the index stream
tikhomirov@242: 					} else {
tikhomirov@420: 						streamOffset = offset;
tikhomirov@242: 						streamDataAccess = daData;
tikhomirov@420: 						daData.longSeek(streamOffset);
tikhomirov@242: 					}
tikhomirov@242: 					final boolean patchToPrevious = baseRevision != i; // the only way I found to tell if it's a patch
tikhomirov@397: 					if (streamDataAccess.isEmpty() || compressedLen == 0) {
tikhomirov@242: 						userDataAccess = new DataAccess(); // empty
tikhomirov@242: 					} else {
tikhomirov@242: 						final byte firstByte = streamDataAccess.readByte();
tikhomirov@242: 						if (firstByte == 0x78 /* 'x' */) {
tikhomirov@263: 							inflater.reset();
tikhomirov@263: 							userDataAccess = new InflaterDataAccess(streamDataAccess, streamOffset, compressedLen, patchToPrevious ? -1 : actualLen, inflater, inflaterBuffer);
tikhomirov@242: 						} else if (firstByte == 0x75 /* 'u' */) {
tikhomirov@242: 							userDataAccess = new FilterDataAccess(streamDataAccess, streamOffset+1, compressedLen-1);
tikhomirov@242: 						} else {
tikhomirov@397: 							// XXX Python impl in fact throws exception when there's not 'x', 'u' or '0' but I don't see reason not to return data as is
tikhomirov@397: 							//
tikhomirov@397: 							// although firstByte is already read from the streamDataAccess, FilterDataAccess#readByte would seek to
tikhomirov@397: 							// initial offset before first attempt to read a byte
tikhomirov@242: 							userDataAccess = new FilterDataAccess(streamDataAccess, streamOffset, compressedLen);
tikhomirov@242: 						}
tikhomirov@242: 					}
tikhomirov@397: 					// userDataAccess is revision content, either complete revision, patch of a previous content, or an empty patch
tikhomirov@397: 					if (patchToPrevious) {
tikhomirov@242: 						// this is a patch
tikhomirov@397: 						if (userDataAccess.isEmpty()) {
tikhomirov@397: 							// Issue 22, empty patch to an empty base revision
tikhomirov@397: 							// Issue 24, empty patch to non-empty base revision
tikhomirov@397: 							// empty patch modifies nothing, use content of a previous revision (shall present - it's a patch here)
tikhomirov@397: 							//
tikhomirov@397: 							assert lastUserData.length() == actualLen; // with no patch, data size shall be the same
tikhomirov@397: 							userDataAccess = lastUserData;
tikhomirov@397: 						} else {
tikhomirov@397: 							patch.read(userDataAccess);
tikhomirov@397: 							userDataAccess.done();
tikhomirov@397: 							//
tikhomirov@397: 							// it shall be reset at the end of prev iteration, when it got assigned from userDataAccess
tikhomirov@397: 							// however, actual userDataAccess and lastUserData may share Inflater object, which needs to be reset
tikhomirov@397: 							// Alternatively, userDataAccess.done() above may be responsible to reset Inflater (if it's InflaterDataAccess)
tikhomirov@397: 							lastUserData.reset();
tikhomirov@397: //							final long startMeasuring = System.currentTimeMillis(); // TIMING
tikhomirov@397: 							byte[] userData = patch.apply(lastUserData, actualLen);
tikhomirov@397: //							applyTime += (System.currentTimeMillis() - startMeasuring); // TIMING
tikhomirov@397: 							patch.clear(); // do not keep any reference, allow byte[] data to be gc'd
tikhomirov@397: 							userDataAccess = new ByteArrayDataAccess(userData);
tikhomirov@397: 						}
tikhomirov@242: 					}
tikhomirov@242: 				} else {
tikhomirov@242: 					if (inline) {
tikhomirov@242: 						daIndex.skip(compressedLen);
tikhomirov@242: 					}
tikhomirov@242: 				}
tikhomirov@329: 				if (i >= start) {
tikhomirov@264: //					final long startMeasuring = System.currentTimeMillis(); // TIMING
tikhomirov@242: 					inspector.next(i, actualLen, baseRevision, linkRevision, parent1Revision, parent2Revision, nodeidBuf, userDataAccess);
tikhomirov@264: //					inspectorTime += (System.currentTimeMillis() - startMeasuring); // TIMING
tikhomirov@242: 				}
tikhomirov@242: 				if (cb != null) {
tikhomirov@242: 					if (cb.isStopped()) {
tikhomirov@242: 						return false;
tikhomirov@242: 					}
tikhomirov@242: 				}
tikhomirov@242: 				if (userDataAccess != null) {
tikhomirov@263: 					userDataAccess.reset(); // not sure this is necessary here, as lastUserData would get reset anyway before next use.
tikhomirov@242: 				}
tikhomirov@397: 				if (lastUserData != null && lastUserData != userDataAccess /* empty patch case, reuse of recent data in actual revision */) {
tikhomirov@397: 					// release lastUserData only if we didn't reuse it in actual revision due to empty patch:
tikhomirov@397: 					// empty patch means we have previous revision and didn't alter it with a patch, hence use lastUserData for userDataAccess above
tikhomirov@263: 					lastUserData.done();
tikhomirov@263: 				}
tikhomirov@263: 				lastUserData = userDataAccess;
tikhomirov@242: 			}
tikhomirov@242: 			lastRevisionRead = end;
tikhomirov@242: 			return true;
tikhomirov@242: 		}
tikhomirov@242: 	}
tikhomirov@242: 
tikhomirov@242: 	
tikhomirov@77: 	public interface Inspector {
tikhomirov@77: 		// XXX boolean retVal to indicate whether to continue?
tikhomirov@157: 		// TODO specify nodeid and data length, and reuse policy (i.e. if revlog stream doesn't reuse nodeid[] for each call)
tikhomirov@157: 		// implementers shall not invoke DataAccess.done(), it's accomplished by #iterate at appropraite moment
tikhomirov@423: 		void next(int revisionIndex, int actualLen, int baseRevision, int linkRevision, int parent1Revision, int parent2Revision, byte[/*20*/] nodeid, DataAccess data);
tikhomirov@3: 	}
tikhomirov@0: }