view src/org/tmatesoft/hg/internal/RevlogStream.java @ 265:3dd953c65619

Generous defaults for SparseSet not to fail on big manifests
author Artem Tikhomirov <tikhomirov.artem@gmail.com>
date Fri, 19 Aug 2011 04:02:48 +0200
parents 6bb5e7ed051a
children 35125450c804
line wrap: on
line source
/*
 * Copyright (c) 2010-2011 TMate Software Ltd
 *  
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; version 2 of the License.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * For information on how to redistribute this software under
 * the terms of a license other than GNU General Public License
 * contact TMate Software at support@hg4j.com
 */
package org.tmatesoft.hg.internal;

import static org.tmatesoft.hg.repo.HgRepository.BAD_REVISION;
import static org.tmatesoft.hg.repo.HgRepository.TIP;

import java.io.File;
import java.io.IOException;
import java.util.ArrayList;
import java.util.List;
import java.util.zip.Inflater;

import org.tmatesoft.hg.core.HgBadStateException;
import org.tmatesoft.hg.core.Nodeid;
import org.tmatesoft.hg.repo.HgRepository;


/**
 * ? Single RevlogStream per file per repository with accessor to record access session (e.g. with back/forward operations), 
 * or numerous RevlogStream with separate representation of the underlying data (cached, lazy ChunkStream)?
 * 
 * @see http://mercurial.selenic.com/wiki/Revlog
 * @see http://mercurial.selenic.com/wiki/RevlogNG
 *
 * @author Artem Tikhomirov
 * @author TMate Software Ltd.
 */
public class RevlogStream {

	/*
	 * makes sense for index with inline data only - actual offset of the record in the .i file (record entry + revision * record size))
	 * 
	 * long[] in fact (there are 8-bytes field in the revlog)
	 * However, (a) DataAccess currently doesn't operate with long seek/length
	 * and, of greater significance, (b) files with inlined data are designated for smaller files,  
	 * guess, about 130 Kb, and offset there won't ever break int capacity
	 */
	private int[] indexRecordOffset;  
	private int[] baseRevisions;
	private boolean inline = false;
	private final File indexFile;
	private final DataAccessProvider dataAccess;

	// if we need anything else from HgRepo, might replace DAP parameter with HgRepo and query it for DAP.
	public RevlogStream(DataAccessProvider dap, File indexFile) {
		this.dataAccess = dap;
		this.indexFile = indexFile;
	}

	/*package*/ DataAccess getIndexStream() {
		return dataAccess.create(indexFile);
	}

	/*package*/ DataAccess getDataStream() {
		final String indexName = indexFile.getName();
		File dataFile = new File(indexFile.getParentFile(), indexName.substring(0, indexName.length() - 1) + "d");
		return dataAccess.create(dataFile);
	}
	
	public int revisionCount() {
		initOutline();
		return baseRevisions.length;
	}
	
	public int dataLength(int revision) {
		// XXX in fact, use of iterate() instead of this implementation may be quite reasonable.
		//
		final int indexSize = revisionCount();
		DataAccess daIndex = getIndexStream(); // XXX may supply a hint that I'll need really few bytes of data (although at some offset)
		if (revision == TIP) {
			revision = indexSize - 1;
		}
		try {
			int recordOffset = getIndexOffsetInt(revision);
			daIndex.seek(recordOffset + 12); // 6+2+4
			int actualLen = daIndex.readInt();
			return actualLen; 
		} catch (IOException ex) {
			ex.printStackTrace(); // log error. FIXME better handling
			throw new IllegalStateException(ex);
		} finally {
			daIndex.done();
		}
	}
	
	public byte[] nodeid(int revision) {
		final int indexSize = revisionCount();
		if (revision == TIP) {
			revision = indexSize - 1;
		}
		if (revision < 0 || revision >= indexSize) {
			throw new IllegalArgumentException(Integer.toString(revision));
		}
		DataAccess daIndex = getIndexStream();
		try {
			int recordOffset = getIndexOffsetInt(revision);
			daIndex.seek(recordOffset + 32);
			byte[] rv = new byte[20];
			daIndex.readBytes(rv, 0, 20);
			return rv;
		} catch (IOException ex) {
			ex.printStackTrace();
			throw new IllegalStateException();
		} finally {
			daIndex.done();
		}
	}
	
	public int linkRevision(int revision) {
		final int last = revisionCount() - 1;
		if (revision == TIP) {
			revision = last;
		}
		if (revision < 0 || revision > last) {
			throw new IllegalArgumentException(Integer.toString(revision));
		}
		DataAccess daIndex = getIndexStream();
		try {
			int recordOffset = getIndexOffsetInt(revision);
			daIndex.seek(recordOffset + 20);
			int linkRev = daIndex.readInt();
			return linkRev;
		} catch (IOException ex) {
			ex.printStackTrace();
			throw new IllegalStateException();
		} finally {
			daIndex.done();
		}
	}
	
	// Perhaps, RevlogStream should be limited to use of plain int revisions for access,
	// while Nodeids should be kept on the level up, in Revlog. Guess, Revlog better keep
	// map of nodeids, and once this comes true, we may get rid of this method.
	// Unlike its counterpart, {@link Revlog#getLocalRevisionNumber()}, doesn't fail with exception if node not found,
	/**
	 * @return integer in [0..revisionCount()) or {@link HgRepository#BAD_REVISION} if not found
	 */
	public int findLocalRevisionNumber(Nodeid nodeid) {
		// XXX this one may be implemented with iterate() once there's mechanism to stop iterations
		final int indexSize = revisionCount();
		DataAccess daIndex = getIndexStream();
		try {
			byte[] nodeidBuf = new byte[20];
			for (int i = 0; i < indexSize; i++) {
				daIndex.skip(8);
				int compressedLen = daIndex.readInt();
				daIndex.skip(20);
				daIndex.readBytes(nodeidBuf, 0, 20);
				if (nodeid.equalsTo(nodeidBuf)) {
					return i;
				}
				daIndex.skip(inline ? 12 + compressedLen : 12);
			}
		} catch (IOException ex) {
			ex.printStackTrace(); // log error. FIXME better handling. Perhaps, shall return BAD_REVISION here as well?
			throw new IllegalStateException(ex);
		} finally {
			daIndex.done();
		}
		return BAD_REVISION;
	}


	private final int REVLOGV1_RECORD_SIZE = 64;

	// should be possible to use TIP, ALL, or -1, -2, -n notation of Hg
	// ? boolean needsNodeid
	public void iterate(int start, int end, boolean needData, Inspector inspector) {
		initOutline();
		final int indexSize = revisionCount();
		if (indexSize == 0) {
			return;
		}
		if (end == TIP) {
			end = indexSize - 1;
		}
		if (start == TIP) {
			start = indexSize - 1;
		}
		if (start < 0 || start >= indexSize) {
			throw new IllegalArgumentException(String.format("Bad left range boundary %d in [0..%d]", start, indexSize-1));
		}
		if (end < start || end >= indexSize) {
			throw new IllegalArgumentException(String.format("Bad right range boundary %d in [0..%d]", end, indexSize-1));
		}
		// XXX may cache [start .. end] from index with a single read (pre-read)
		
		ReaderN1 r = new ReaderN1(needData, inspector);
		try {
			r.start(end - start + 1);
			r.range(start, end);
		} catch (IOException ex) {
			throw new HgBadStateException(ex); // FIXME need better handling
		} finally {
			r.finish();
		}
	}
	
	/**
	 * Effective alternative to {@link #iterate(int, int, boolean, Inspector) batch read}, when only few selected 
	 * revisions are of interest.
	 * @param sortedRevisions revisions to walk, in ascending order.
	 * @param needData whether inspector needs access to header only
	 * @param inspector callback to process entries
	 */
	public void iterate(int[] sortedRevisions, boolean needData, Inspector inspector) {
		final int indexSize = revisionCount();
		if (indexSize == 0 || sortedRevisions.length == 0) {
			return;
		}
		if (sortedRevisions[0] > indexSize || sortedRevisions[sortedRevisions.length - 1] > indexSize) {
			throw new IllegalArgumentException(String.format("Can't iterate [%d, %d] in range [0..%d]", sortedRevisions[0], sortedRevisions[sortedRevisions.length - 1], indexSize));
		}

		ReaderN1 r = new ReaderN1(needData, inspector);
		try {
			r.start(sortedRevisions.length);
			for (int i = 0; i < sortedRevisions.length; ) {
				int x = i;
				i++;
				while (i < sortedRevisions.length) {
					if (sortedRevisions[i] == sortedRevisions[i-1] + 1) {
						i++;
					} else {
						break;
					}
				}
				// commitRevisions[x..i-1] are sequential
				if (!r.range(sortedRevisions[x], sortedRevisions[i-1])) {
					return;
				}
			}
		} catch (IOException ex) {
			throw new HgBadStateException(ex); // FIXME need better handling
		} finally {
			r.finish();
		}
	}

	private int getBaseRevision(int revision) {
		return baseRevisions[revision];
	}

	/**
	 * @return  offset of the revision's record in the index (.i) stream
	 */
	private int getIndexOffsetInt(int revision) {
		return inline ? indexRecordOffset[revision] : revision * REVLOGV1_RECORD_SIZE;
	}

	private void initOutline() {
		if (baseRevisions != null && baseRevisions.length > 0) {
			return;
		}
		ArrayList<Integer> resBases = new ArrayList<Integer>();
		ArrayList<Integer> resOffsets = new ArrayList<Integer>();
		DataAccess da = getIndexStream();
		try {
			if (da.isEmpty()) {
				// do not fail with exception if stream is empty, it's likely intentional
				baseRevisions = new int[0];
				return;
			}
			int versionField = da.readInt();
			da.readInt(); // just to skip next 4 bytes of offset + flags
			final int INLINEDATA = 1 << 16;
			inline = (versionField & INLINEDATA) != 0;
			long offset = 0; // first offset is always 0, thus Hg uses it for other purposes
			while(true) {
				int compressedLen = da.readInt();
				// 8+4 = 12 bytes total read here
				@SuppressWarnings("unused")
				int actualLen = da.readInt();
				int baseRevision = da.readInt();
				// 12 + 8 = 20 bytes read here
//				int linkRevision = di.readInt();
//				int parent1Revision = di.readInt();
//				int parent2Revision = di.readInt();
//				byte[] nodeid = new byte[32];
				resBases.add(baseRevision);
				if (inline) {
					int o = (int) offset;
					if (o != offset) {
						// just in case, can't happen, ever, unless HG (or some other bad tool) produces index file 
						// with inlined data of size greater than 2 Gb.
						throw new HgBadStateException("Data too big, offset didn't fit to sizeof(int)");
					}
					resOffsets.add(o + REVLOGV1_RECORD_SIZE * resOffsets.size());
					da.skip(3*4 + 32 + compressedLen); // Check: 44 (skip) + 20 (read) = 64 (total RevlogNG record size)
				} else {
					da.skip(3*4 + 32);
				}
				if (da.isEmpty()) {
					// fine, done then
					baseRevisions = toArray(resBases);
					if (inline) {
						indexRecordOffset = toArray(resOffsets);
					}
					break;
				} else {
					// start reading next record
					long l = da.readLong();
					offset = l >>> 16;
				}
			}
		} catch (IOException ex) {
			ex.printStackTrace(); // log error
			// too bad, no outline then, but don't fail with NPE
			baseRevisions = new int[0];
		} finally {
			da.done();
		}
	}
	
	/**
	 * operation with single file open/close and multiple diverse reads.
	 * XXX initOutline might need similar extraction to keen N1 format knowledge  
	 */
	class ReaderN1 {
		private final Inspector inspector;
		private final boolean needData;
		private DataAccess daIndex = null, daData = null;
		private Lifecycle.BasicCallback cb = null;
		private int lastRevisionRead = BAD_REVISION;
		private DataAccess lastUserData;
		// next are to track two major bottlenecks - patch application and actual time spent in inspector 
//		private long applyTime, inspectorTime; // TIMING


		public ReaderN1(boolean needData, Inspector insp) {
			assert insp != null;
			this.needData = needData;
			inspector = insp;
		}
		
		public void start(int totalWork) {
			daIndex = getIndexStream();
			if (needData && !inline) {
				daData = getDataStream();
			}
			if (inspector instanceof Lifecycle) {
				cb = new Lifecycle.BasicCallback();
				((Lifecycle) inspector).start(totalWork, cb, cb);
			}
//			applyTime = inspectorTime = 0; // TIMING
		}

		public void finish() {
			if (lastUserData != null) {
				lastUserData.done();
				lastUserData = null;
			}
			if (inspector instanceof Lifecycle) {
				((Lifecycle) inspector).finish(cb);
			}
			daIndex.done();
			if (daData != null) {
				daData.done();
			}
//			System.out.printf("applyTime:%d ms, inspectorTime: %d ms\n", applyTime, inspectorTime); // TIMING
		}

		public boolean range(int start, int end) throws IOException {
			byte[] nodeidBuf = new byte[20];
			int i;
			boolean extraReadsToBaseRev = false; // to indicate we read revision prior to start. XXX not sure can't do without
			// it (i.e. replace with i >= start)
			if (needData && (i = getBaseRevision(start)) < start) {
				// if lastRevisionRead in [baseRevision(start), start)  can reuse lastUserData
				// doesn't make sense to reuse if lastRevisionRead == start (too much to change in the cycle below). 
				if (lastRevisionRead != BAD_REVISION && i <= lastRevisionRead && lastRevisionRead < start) {
					i = lastRevisionRead + 1; // start with first not-yet-read revision
					extraReadsToBaseRev = i < start;
				} else {
					if (lastUserData != null) {
						lastUserData.done();
						lastUserData = null;
					}
					extraReadsToBaseRev = true;
				}
			} else {
				// don't need to clean lastUserData as it's always null when !needData
				i = start;
			}
			
			daIndex.seek(getIndexOffsetInt(i));
			//
			// reuse some instances
			final ArrayList<PatchRecord> patches = new ArrayList<PatchRecord>();
			final Inflater inflater = new Inflater();
			// can share buffer between instances of InflaterDataAccess as I never read any two of them in parallel
			final byte[] inflaterBuffer = new byte[1024];
			//
			
			for (; i <= end; i++ ) {
				if (inline && needData) {
					// inspector reading data (though FilterDataAccess) may have affected index position
					daIndex.seek(getIndexOffsetInt(i));
				}
				long l = daIndex.readLong(); // 0
				long offset = i == 0 ? 0 : (l >>> 16);
				@SuppressWarnings("unused")
				int flags = (int) (l & 0X0FFFF);
				int compressedLen = daIndex.readInt(); // +8
				int actualLen = daIndex.readInt(); // +12
				int baseRevision = daIndex.readInt(); // +16
				int linkRevision = daIndex.readInt(); // +20
				int parent1Revision = daIndex.readInt();
				int parent2Revision = daIndex.readInt();
				// Hg has 32 bytes here, uses 20 for nodeid, and keeps 12 last bytes empty
				daIndex.readBytes(nodeidBuf, 0, 20); // +32
				daIndex.skip(12);
				DataAccess userDataAccess = null;
				if (needData) {
					int streamOffset;
					DataAccess streamDataAccess;
					if (inline) {
						streamDataAccess = daIndex;
						streamOffset = getIndexOffsetInt(i) + REVLOGV1_RECORD_SIZE; // don't need to do seek as it's actual position in the index stream
					} else {
						streamOffset = (int) offset;
						streamDataAccess = daData;
						daData.seek(streamOffset);
					}
					final boolean patchToPrevious = baseRevision != i; // the only way I found to tell if it's a patch
					if (streamDataAccess.isEmpty()) {
						userDataAccess = new DataAccess(); // empty
					} else {
						final byte firstByte = streamDataAccess.readByte();
						if (firstByte == 0x78 /* 'x' */) {
							inflater.reset();
							userDataAccess = new InflaterDataAccess(streamDataAccess, streamOffset, compressedLen, patchToPrevious ? -1 : actualLen, inflater, inflaterBuffer);
						} else if (firstByte == 0x75 /* 'u' */) {
							userDataAccess = new FilterDataAccess(streamDataAccess, streamOffset+1, compressedLen-1);
						} else {
							// 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 
							userDataAccess = new FilterDataAccess(streamDataAccess, streamOffset, compressedLen);
						}
					}
					// XXX 
					if (patchToPrevious) {
						// this is a patch
						patches.clear(); // won't hurt to ensure there are no leftovers, even if we already cleaned
						while (!userDataAccess.isEmpty()) {
							PatchRecord pr = PatchRecord.read(userDataAccess);
//							System.out.printf("PatchRecord:%d %d %d\n", pr.start, pr.end, pr.len);
							patches.add(pr);
						}
						userDataAccess.done();
						//
						// it shall be reset at the end of prev iteration, when it got assigned from userDataAccess
						// however, actual userDataAccess and lastUserData may share Inflater object, which needs to be reset
						// Alternatively, userDataAccess.done() above may be responsible to reset Inflater (if it's InflaterDataAccess)
						lastUserData.reset();
//						final long startMeasuring = System.currentTimeMillis(); // TIMING
						byte[] userData = apply(lastUserData, actualLen, patches);
//						applyTime += (System.currentTimeMillis() - startMeasuring); // TIMING
						patches.clear(); // do not keep any reference, allow PatchRecord to be gc'd
						userDataAccess = new ByteArrayDataAccess(userData);
					}
				} else {
					if (inline) {
						daIndex.skip(compressedLen);
					}
				}
				if (!extraReadsToBaseRev || i >= start) {
//					final long startMeasuring = System.currentTimeMillis(); // TIMING
					inspector.next(i, actualLen, baseRevision, linkRevision, parent1Revision, parent2Revision, nodeidBuf, userDataAccess);
//					inspectorTime += (System.currentTimeMillis() - startMeasuring); // TIMING
				}
				if (cb != null) {
					if (cb.isStopped()) {
						return false;
					}
				}
				if (userDataAccess != null) {
					userDataAccess.reset(); // not sure this is necessary here, as lastUserData would get reset anyway before next use.
				}
				if (lastUserData != null) {
					lastUserData.done();
				}
				lastUserData = userDataAccess;
			}
			lastRevisionRead = end;
			return true;
		}
	}

	
	private static int[] toArray(List<Integer> l) {
		int[] rv = new int[l.size()];
		for (int i = 0; i < rv.length; i++) {
			rv[i] = l.get(i);
		}
		return rv;
	}
	

	// mpatch.c : apply()
	// FIXME need to implement patch merge (fold, combine, gather and discard from aforementioned mpatch.[c|py]), also see Revlog and Mercurial PDF
	public/*for HgBundle; until moved to better place*/static byte[] apply(DataAccess baseRevisionContent, int outcomeLen, List<PatchRecord> patch) throws IOException {
		int last = 0, destIndex = 0;
		if (outcomeLen == -1) {
			outcomeLen = baseRevisionContent.length();
			for (int i = 0, x = patch.size(); i < x; i++) {
				PatchRecord pr = patch.get(i);
				outcomeLen += pr.start - last + pr.len;
				last = pr.end;
			}
			outcomeLen -= last;
			last = 0;
		}
		byte[] rv = new byte[outcomeLen];
		for (int i = 0, x = patch.size(); i < x; i++) {
			PatchRecord pr = patch.get(i);
			baseRevisionContent.seek(last);
			baseRevisionContent.readBytes(rv, destIndex, pr.start-last);
			destIndex += pr.start - last;
			System.arraycopy(pr.data, 0, rv, destIndex, pr.data.length);
			destIndex += pr.data.length;
			last = pr.end;
		}
		baseRevisionContent.seek(last);
		baseRevisionContent.readBytes(rv, destIndex, (int) (baseRevisionContent.length() - last));
		return rv;
	}

	// @see http://mercurial.selenic.com/wiki/BundleFormat, in Changelog group description
	public static class PatchRecord {
		/*
		   Given there are pr1 and pr2:
		     pr1.start to pr1.end will be replaced with pr's data (of pr1.len)
		     pr1.end to pr2.start gets copied from base
		 */
		public int start, end, len;
		public byte[] data;

		// TODO consider PatchRecord that only records data position (absolute in data source), and acquires data as needed 
		private PatchRecord(int p1, int p2, int length, byte[] src) {
			start = p1;
			end = p2;
			len = length;
			data = src;
		}

		/*package-local*/ static PatchRecord read(byte[] data, int offset) {
			final int x = offset; // shorthand
			int p1 =  ((data[x] & 0xFF)<< 24)    | ((data[x+1] & 0xFF) << 16) | ((data[x+2] & 0xFF) << 8)  | (data[x+3] & 0xFF);
			int p2 =  ((data[x+4] & 0xFF) << 24) | ((data[x+5] & 0xFF) << 16) | ((data[x+6] & 0xFF) << 8)  | (data[x+7] & 0xFF);
			int len = ((data[x+8] & 0xFF) << 24) | ((data[x+9] & 0xFF) << 16) | ((data[x+10] & 0xFF) << 8) | (data[x+11] & 0xFF);
			byte[] dataCopy = new byte[len];
			System.arraycopy(data, x+12, dataCopy, 0, len);
			return new PatchRecord(p1, p2, len, dataCopy);
		}

		public /*for HgBundle*/ static PatchRecord read(DataAccess da) throws IOException {
			int p1 = da.readInt();
			int p2 = da.readInt();
			int len = da.readInt();
			byte[] src = new byte[len];
			da.readBytes(src, 0, len);
			return new PatchRecord(p1, p2, len, src);
		}
	}

	// FIXME byte[] data might be too expensive, for few usecases it may be better to have intermediate Access object (when we don't need full data 
	// instantly - e.g. calculate hash, or comparing two revisions
	public interface Inspector {
		// XXX boolean retVal to indicate whether to continue?
		// TODO specify nodeid and data length, and reuse policy (i.e. if revlog stream doesn't reuse nodeid[] for each call)
		// implementers shall not invoke DataAccess.done(), it's accomplished by #iterate at appropraite moment
		void next(int revisionNumber, int actualLen, int baseRevision, int linkRevision, int parent1Revision, int parent2Revision, byte[/*20*/] nodeid, DataAccess data);
	}
}