Mercurial > jhg
view src/org/tmatesoft/hg/repo/HgBlameFacility.java @ 580:bd5926e24aa3
Respect unix flags for checkout/revert
author | Artem Tikhomirov <tikhomirov.artem@gmail.com> |
---|---|
date | Fri, 19 Apr 2013 20:30:34 +0200 |
parents | e49f9d9513fa |
children | 43cfa08ff3fd |
line wrap: on
line source
/* * Copyright (c) 2013 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.repo; import static org.tmatesoft.hg.core.HgIterateDirection.NewToOld; import static org.tmatesoft.hg.core.HgIterateDirection.OldToNew; import static org.tmatesoft.hg.repo.HgInternals.wrongRevisionIndex; import static org.tmatesoft.hg.repo.HgRepository.*; import java.util.Arrays; import java.util.BitSet; import java.util.Collections; import java.util.LinkedList; import org.tmatesoft.hg.core.HgCallbackTargetException; import org.tmatesoft.hg.core.HgIterateDirection; import org.tmatesoft.hg.core.Nodeid; import org.tmatesoft.hg.internal.BlameHelper; import org.tmatesoft.hg.internal.Callback; import org.tmatesoft.hg.internal.Experimental; import org.tmatesoft.hg.internal.IntVector; import org.tmatesoft.hg.util.Adaptable; /** * Facility with diff/annotate functionality. * * @author Artem Tikhomirov * @author TMate Software Ltd. */ @Experimental(reason="Unstable API") public final class HgBlameFacility { private final HgDataFile df; public HgBlameFacility(HgDataFile file) { if (file == null) { throw new IllegalArgumentException(); } df = file; } /** * mimic 'hg diff -r clogRevIndex1 -r clogRevIndex2' */ public void diff(int clogRevIndex1, int clogRevIndex2, Inspector insp) throws HgCallbackTargetException { // FIXME clogRevIndex1 and clogRevIndex2 may point to different files, need to decide whether to throw an exception // or to attempt to look up correct file node (tricky) int fileRevIndex1 = fileRevIndex(df, clogRevIndex1); int fileRevIndex2 = fileRevIndex(df, clogRevIndex2); BlameHelper bh = new BlameHelper(insp, 5); bh.useFileUpTo(df, clogRevIndex2); bh.diff(fileRevIndex1, clogRevIndex1, fileRevIndex2, clogRevIndex2); } /** * Walk file history up/down to revision at given changeset and report changes for each revision */ public void annotate(int changelogRevisionIndex, Inspector insp, HgIterateDirection iterateOrder) throws HgCallbackTargetException { annotate(0, changelogRevisionIndex, insp, iterateOrder); } /** * Walk file history range and report changes for each revision */ public void annotate(int changelogRevIndexStart, int changelogRevIndexEnd, Inspector insp, HgIterateDirection iterateOrder) throws HgCallbackTargetException { if (wrongRevisionIndex(changelogRevIndexStart) || wrongRevisionIndex(changelogRevIndexEnd)) { throw new IllegalArgumentException(); } // Note, changelogRevIndexEnd may be TIP, while the code below doesn't tolerate constants // int lastRevision = df.getRepo().getChangelog().getLastRevision(); if (changelogRevIndexEnd == TIP) { changelogRevIndexEnd = lastRevision; } HgInternals.checkRevlogRange(changelogRevIndexStart, changelogRevIndexEnd, lastRevision); if (!df.exists()) { return; } BlameHelper bh = new BlameHelper(insp, 10); HgDataFile currentFile = df; int fileLastClogRevIndex = changelogRevIndexEnd; FileRevisionHistoryChunk nextChunk = null; LinkedList<FileRevisionHistoryChunk> fileCompleteHistory = new LinkedList<FileRevisionHistoryChunk>(); do { FileRevisionHistoryChunk fileHistory = new FileRevisionHistoryChunk(currentFile); fileHistory.init(fileLastClogRevIndex); fileHistory.linkTo(nextChunk); fileCompleteHistory.addFirst(fileHistory); // to get the list in old-to-new order nextChunk = fileHistory; bh.useFileUpTo(currentFile, fileLastClogRevIndex); if (fileHistory.changeset(0) > changelogRevIndexStart && currentFile.isCopy()) { // fileHistory.changeset(0) is the earliest revision we know about so far, // once we get to revisions earlier than the requested start, stop digging. // The reason there's NO == (i.e. not >=) because: // (easy): once it's equal, we've reached our intended start // (hard): if changelogRevIndexStart happens to be exact start of one of renames in the // chain of renames (test-annotate2 repository, file1->file1a->file1b, i.e. points // to the very start of file1a or file1 history), presence of == would get us to the next // chunk and hence changed parents of present chunk's first element. Our annotate alg // relies on parents only (i.e. knows nothing about 'last iteration element') to find out // what to compare, and hence won't report all lines of 'last iteration element' (which is the // first revision of the renamed file) as "added in this revision", leaving gaps in annotate HgRepository repo = currentFile.getRepo(); Nodeid originLastRev = currentFile.getCopySourceRevision(); currentFile = repo.getFileNode(currentFile.getCopySourceName()); fileLastClogRevIndex = currentFile.getChangesetRevisionIndex(currentFile.getRevisionIndex(originLastRev)); // XXX perhaps, shall fail with meaningful exception if new file doesn't exist (.i/.d not found for whatever reason) // or source revision is missing? } else { fileHistory.chopAtChangeset(changelogRevIndexStart); currentFile = null; // stop iterating } } while (currentFile != null && fileLastClogRevIndex > changelogRevIndexStart); // fileCompleteHistory is in (origin, intermediate target, ultimate target) order int[] fileClogParentRevs = new int[2]; int[] fileParentRevs = new int[2]; if (iterateOrder == NewToOld) { Collections.reverse(fileCompleteHistory); } boolean shallFilterStart = changelogRevIndexStart != 0; // no reason if complete history is walked for (FileRevisionHistoryChunk fileHistory : fileCompleteHistory) { for (int fri : fileHistory.fileRevisions(iterateOrder)) { int clogRevIndex = fileHistory.changeset(fri); if (shallFilterStart) { if (iterateOrder == NewToOld) { // clogRevIndex decreases if (clogRevIndex < changelogRevIndexStart) { break; } // fall-through, clogRevIndex is in the [start..end] range } else { // old to new // the way we built fileHistory ensures we won't walk past changelogRevIndexEnd // here we ensure we start from the right one, the one indicated with changelogRevIndexStart if (clogRevIndex < changelogRevIndexStart) { continue; } else { shallFilterStart = false; // once boundary is crossed, no need to check // fall-through } } } fileHistory.fillFileParents(fri, fileParentRevs); fileHistory.fillCsetParents(fri, fileClogParentRevs); bh.annotateChange(fri, clogRevIndex, fileParentRevs, fileClogParentRevs); } } } /** * Annotates changes of the file against its parent(s). * Unlike {@link #annotate(HgDataFile, int, Inspector, HgIterateDirection)}, doesn't * walk file history, looks at the specified revision only. Handles both parents (if merge revision). */ public void annotateSingleRevision(int changelogRevisionIndex, Inspector insp) throws HgCallbackTargetException { // TODO detect if file is text/binary (e.g. looking for chars < ' ' and not \t\r\n\f int fileRevIndex = fileRevIndex(df, changelogRevisionIndex); int[] fileRevParents = new int[2]; df.parents(fileRevIndex, fileRevParents, null, null); if (changelogRevisionIndex == TIP) { changelogRevisionIndex = df.getChangesetRevisionIndex(fileRevIndex); } BlameHelper bh = new BlameHelper(insp, 5); bh.useFileUpTo(df, changelogRevisionIndex); int[] fileClogParentRevs = new int[2]; fileClogParentRevs[0] = fileRevParents[0] == NO_REVISION ? NO_REVISION : df.getChangesetRevisionIndex(fileRevParents[0]); fileClogParentRevs[1] = fileRevParents[1] == NO_REVISION ? NO_REVISION : df.getChangesetRevisionIndex(fileRevParents[1]); bh.annotateChange(fileRevIndex, changelogRevisionIndex, fileRevParents, fileClogParentRevs); } /** * Client's sink for revision differences. * * When implemented, clients shall not expect new {@link Block blocks} instances in each call. * * In case more information about annotated revision is needed, inspector instances may supply * {@link RevisionDescriptor.Recipient} through {@link Adaptable}. */ @Callback public interface Inspector { void same(EqualBlock block) throws HgCallbackTargetException; void added(AddBlock block) throws HgCallbackTargetException; void changed(ChangeBlock block) throws HgCallbackTargetException; void deleted(DeleteBlock block) throws HgCallbackTargetException; } /** * No need to keep "Block" prefix as long as there's only one {@link Inspector} */ @Deprecated public interface BlockInspector extends Inspector { } /** * Represents content of a block, either as a sequence of bytes or a * sequence of smaller blocks (lines), if appropriate (according to usage context). * * This approach allows line-by-line access to content data along with complete byte sequence for the whole block, i.e. * <pre> * BlockData bd = addBlock.addedLines() * // bd describes data from the addition completely. * // elements of the BlockData are lines * bd.elementCount() == addBlock.totalAddedLines(); * // one cat obtain complete addition with * byte[] everythingAdded = bd.asArray(); * // or iterate line by line * for (int i = 0; i < bd.elementCount(); i++) { * byte[] lineContent = bd.elementAt(i); * String line = new String(lineContent, fileEncodingCharset); * } * where bd.elementAt(0) is the line at index addBlock.firstAddedLine() * </pre> * * LineData or ChunkData? */ public interface BlockData { BlockData elementAt(int index); int elementCount(); byte[] asArray(); } /** * {@link Inspector} may optionally request extra information about revisions * being inspected, denoting itself as a {@link RevisionDescriptor.Recipient}. This class * provides complete information about file revision under annotation now. */ public interface RevisionDescriptor { /** * @return complete source of the diff origin, never <code>null</code> */ BlockData origin(); /** * @return complete source of the diff target, never <code>null</code> */ BlockData target(); /** * @return changeset revision index of original file, or {@link HgRepository#NO_REVISION} if it's the very first revision */ int originChangesetIndex(); /** * @return changeset revision index of the target file */ int targetChangesetIndex(); /** * @return <code>true</code> if this revision is merge */ boolean isMerge(); /** * @return changeset revision index of the second, merged parent */ int mergeChangesetIndex(); /** * @return revision index of the change in target file's revlog */ int fileRevisionIndex(); /** * @return file object under blame (target file) */ HgDataFile file(); /** * Implement to indicate interest in {@link RevisionDescriptor}. * * Note, instance of {@link RevisionDescriptor} is the same for * {@link #start(RevisionDescriptor)} and {@link #done(RevisionDescriptor)} * methods, and not necessarily a new one (i.e. <code>==</code>) for the next * revision announced. */ @Callback public interface Recipient { /** * Comes prior to any change {@link Block blocks} */ void start(RevisionDescriptor revisionDescription) throws HgCallbackTargetException; /** * Comes after all change {@link Block blocks} were dispatched */ void done(RevisionDescriptor revisionDescription) throws HgCallbackTargetException; } } /** * Each change block comes from a single origin, blocks that are result of a merge * have {@link #originChangesetIndex()} equal to {@link RevisionDescriptor#mergeChangesetIndex()}. */ public interface Block { int originChangesetIndex(); int targetChangesetIndex(); } public interface EqualBlock extends Block { int originStart(); int targetStart(); int length(); BlockData content(); } public interface AddBlock extends Block { /** * @return line index in the origin where this block is inserted */ int insertedAt(); /** * @return line index of the first added line in the target revision */ int firstAddedLine(); /** * @return number of added lines in this block */ int totalAddedLines(); /** * @return content of added lines */ BlockData addedLines(); } public interface DeleteBlock extends Block { /** * @return line index in the target revision were this deleted block would be */ int removedAt(); /** * @return line index of the first removed line in the original revision */ int firstRemovedLine(); /** * @return number of deleted lines in this block */ int totalRemovedLines(); /** * @return content of deleted lines */ BlockData removedLines(); } public interface ChangeBlock extends AddBlock, DeleteBlock { } private static int fileRevIndex(HgDataFile df, int csetRevIndex) { Nodeid fileRev = df.getRepo().getManifest().getFileRevision(csetRevIndex, df.getPath()); return df.getRevisionIndex(fileRev); } private static class FileRevisionHistoryChunk { private final HgDataFile df; // change ancestry, sequence of file revisions private IntVector fileRevsToVisit; // parent pairs of complete file history private IntVector fileParentRevs; // map file revision to changelog revision (sparse array, only file revisions to visit are set) private int[] file2changelog; private int originChangelogRev = BAD_REVISION, originFileRev = BAD_REVISION; public FileRevisionHistoryChunk(HgDataFile file) { df = file; } public void init(int changelogRevisionIndex) { // XXX df.indexWalk(0, fileRevIndex, ) might be more effective int fileRevIndex = fileRevIndex(df, changelogRevisionIndex); int[] fileRevParents = new int[2]; fileParentRevs = new IntVector((fileRevIndex+1) * 2, 0); fileParentRevs.add(NO_REVISION, NO_REVISION); // parents of fileRevIndex == 0 for (int i = 1; i <= fileRevIndex; i++) { df.parents(i, fileRevParents, null, null); fileParentRevs.add(fileRevParents[0], fileRevParents[1]); } // fileRevsToVisit keep file change ancestry from new to old fileRevsToVisit = new IntVector(fileRevIndex + 1, 0); // keep map of file revision to changelog revision file2changelog = new int[fileRevIndex+1]; // only elements worth visit would get mapped, so there would be unfilled areas in the file2changelog, // prevent from error (make it explicit) by bad value Arrays.fill(file2changelog, BAD_REVISION); LinkedList<Integer> queue = new LinkedList<Integer>(); BitSet seen = new BitSet(fileRevIndex + 1); queue.add(fileRevIndex); do { int x = queue.removeFirst(); if (seen.get(x)) { continue; } seen.set(x); fileRevsToVisit.add(x); file2changelog[x] = df.getChangesetRevisionIndex(x); int p1 = fileParentRevs.get(2*x); int p2 = fileParentRevs.get(2*x + 1); if (p1 != NO_REVISION) { queue.addLast(p1); } if (p2 != NO_REVISION) { queue.addLast(p2); } } while (!queue.isEmpty()); // make sure no child is processed before we handled all (grand-)parents of the element fileRevsToVisit.sort(false); } public void linkTo(FileRevisionHistoryChunk target) { // assume that target.init() has been called already if (target == null) { return; } target.originFileRev = fileRevsToVisit.get(0); // files to visit are new to old target.originChangelogRev = changeset(target.originFileRev); } /** * Mark revision closest(ceil) to specified as the very first one (no parents) */ public void chopAtChangeset(int firstChangelogRevOfInterest) { if (firstChangelogRevOfInterest == 0) { return; // nothing to do } int i = 0, x = fileRevsToVisit.size(), fileRev = BAD_REVISION; // fileRevsToVisit is new to old, greater numbers to smaller while (i < x && changeset(fileRev = fileRevsToVisit.get(i)) >= firstChangelogRevOfInterest) { i++; } assert fileRev != BAD_REVISION; // there's at least 1 revision in fileRevsToVisit if (i == x && changeset(fileRev) != firstChangelogRevOfInterest) { assert false : "Requested changeset shall belong to the chunk"; return; } fileRevsToVisit.trimTo(i); // no need to iterate more // pretend fileRev got no parents fileParentRevs.set(fileRev * 2, NO_REVISION); fileParentRevs.set(fileRev, NO_REVISION); } public int[] fileRevisions(HgIterateDirection iterateOrder) { // fileRevsToVisit is { r10, r7, r6, r5, r0 }, new to old int[] rv = fileRevsToVisit.toArray(); if (iterateOrder == OldToNew) { // reverse return value for (int a = 0, b = rv.length-1; a < b; a++, b--) { int t = rv[b]; rv[b] = rv[a]; rv[a] = t; } } return rv; } public int changeset(int fileRevIndex) { return file2changelog[fileRevIndex]; } public void fillFileParents(int fileRevIndex, int[] fileParents) { if (fileRevIndex == 0 && originFileRev != BAD_REVISION) { // this chunk continues another file assert originFileRev != NO_REVISION; fileParents[0] = originFileRev; fileParents[1] = NO_REVISION; return; } fileParents[0] = fileParentRevs.get(fileRevIndex * 2); fileParents[1] = fileParentRevs.get(fileRevIndex * 2 + 1); } public void fillCsetParents(int fileRevIndex, int[] csetParents) { if (fileRevIndex == 0 && originFileRev != BAD_REVISION) { assert originFileRev != NO_REVISION; csetParents[0] = originChangelogRev; csetParents[1] = NO_REVISION; // I wonder if possible to start a copy with two parents? return; } int fp1 = fileParentRevs.get(fileRevIndex * 2); int fp2 = fileParentRevs.get(fileRevIndex * 2 + 1); csetParents[0] = fp1 == NO_REVISION ? NO_REVISION : changeset(fp1); csetParents[1] = fp2 == NO_REVISION ? NO_REVISION : changeset(fp2); } } }