Mercurial > jhg
view src/org/tmatesoft/hg/internal/RepositoryComparator.java @ 583:47dfa0ec7e35
Effective revlog patching
author | Artem Tikhomirov <tikhomirov.artem@gmail.com> |
---|---|
date | Wed, 24 Apr 2013 15:39:53 +0200 |
parents | 1fc0da631200 |
children | 6526d8adbc0f |
line wrap: on
line source
/* * Copyright (c) 2011-2012 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.core.Nodeid.NULL; import java.util.ArrayList; import java.util.Arrays; import java.util.Collections; import java.util.HashMap; import java.util.HashSet; import java.util.LinkedList; import java.util.List; import java.util.ListIterator; import java.util.Map; import java.util.Map.Entry; import java.util.Set; import org.tmatesoft.hg.core.HgRemoteConnectionException; import org.tmatesoft.hg.core.Nodeid; import org.tmatesoft.hg.repo.HgChangelog; import org.tmatesoft.hg.repo.HgInvalidControlFileException; import org.tmatesoft.hg.repo.HgInvalidStateException; import org.tmatesoft.hg.repo.HgRemoteRepository; import org.tmatesoft.hg.repo.HgRemoteRepository.Range; import org.tmatesoft.hg.repo.HgRemoteRepository.RemoteBranch; import org.tmatesoft.hg.repo.HgParentChildMap; import org.tmatesoft.hg.util.CancelSupport; import org.tmatesoft.hg.util.CancelledException; import org.tmatesoft.hg.util.ProgressSupport; /** * * @author Artem Tikhomirov * @author TMate Software Ltd. */ public class RepositoryComparator { private final boolean debug = Boolean.parseBoolean(System.getProperty("hg4j.remote.debug")); private final HgParentChildMap<HgChangelog> localRepo; private final HgRemoteRepository remoteRepo; private List<Nodeid> common; public RepositoryComparator(HgParentChildMap<HgChangelog> pwLocal, HgRemoteRepository hgRemote) { localRepo = pwLocal; remoteRepo = hgRemote; } public RepositoryComparator compare(ProgressSupport progressSupport, CancelSupport cancelSupport) throws HgRemoteConnectionException, CancelledException { cancelSupport.checkCancelled(); progressSupport.start(10); common = Collections.unmodifiableList(findCommonWithRemote()); // sanity check for (Nodeid n : common) { if (!localRepo.knownNode(n)) { throw new HgInvalidStateException("Unknown node reported as common:" + n); } } progressSupport.done(); return this; } public List<Nodeid> getCommon() { if (common == null) { throw new HgInvalidStateException("Call #compare(Object) first"); } return common; } /** * @return revisions that are children of common entries, i.e. revisions that are present on the local server and not on remote. */ public List<Nodeid> getLocalOnlyRevisions() { return localRepo.childrenOf(getCommon()); } /** * Similar to @link {@link #getLocalOnlyRevisions()}, use this one if you need access to changelog entry content, not * only its revision number. * @param inspector delegate to analyze changesets, shall not be <code>null</code> */ public void visitLocalOnlyRevisions(HgChangelog.Inspector inspector) throws HgInvalidControlFileException { if (inspector == null) { throw new IllegalArgumentException(); } // one can use localRepo.childrenOf(getCommon()) and then iterate over nodeids, but there seems to be // another approach to get all changes after common: // find index of earliest revision, and report all that were later final HgChangelog changelog = localRepo.getRepo().getChangelog(); int earliestRevision = Integer.MAX_VALUE; List<Nodeid> commonKnown = getCommon(); for (Nodeid n : commonKnown) { if (!localRepo.hasChildren(n)) { // there might be (old) nodes, known both locally and remotely, with no children // hence, we don't need to consider their local revision number continue; } int lr = changelog.getRevisionIndex(n); if (lr < earliestRevision) { earliestRevision = lr; } } if (earliestRevision == Integer.MAX_VALUE) { // either there are no common nodes (known locally and at remote) // or no local children found (local is up to date). In former case, perhaps I shall bit return silently, // but check for possible wrong repo comparison (hs says 'repository is unrelated' if I try to // check in/out for a repo that has no common nodes. return; } if (earliestRevision < 0 || earliestRevision >= changelog.getLastRevision()) { throw new HgInvalidStateException(String.format("Invalid index of common known revision: %d in total of %d", earliestRevision, 1+changelog.getLastRevision())); } changelog.range(earliestRevision+1, changelog.getLastRevision(), inspector); } private List<Nodeid> findCommonWithRemote() throws HgRemoteConnectionException { List<Nodeid> remoteHeads = remoteRepo.heads(); LinkedList<Nodeid> resultCommon = new LinkedList<Nodeid>(); // these remotes are known in local LinkedList<Nodeid> toQuery = new LinkedList<Nodeid>(); // these need further queries to find common for (Nodeid rh : remoteHeads) { if (localRepo.knownNode(rh)) { resultCommon.add(rh); } else { toQuery.add(rh); } } if (toQuery.isEmpty()) { return resultCommon; } LinkedList<RemoteBranch> checkUp2Head = new LinkedList<RemoteBranch>(); // branch.root and branch.head are of interest only. // these are branches with unknown head but known root, which might not be the last common known, // i.e. there might be children changeset that are also available at remote, [..?..common-head..remote-head] - need to // scroll up to common head. while (!toQuery.isEmpty()) { List<RemoteBranch> remoteBranches = remoteRepo.branches(toQuery); //head, root, first parent, second parent toQuery.clear(); while(!remoteBranches.isEmpty()) { RemoteBranch rb = remoteBranches.remove(0); // I assume branches remote call gives branches with head equal to what I pass there, i.e. // that I don't need to check whether rb.head is unknown. if (localRepo.knownNode(rb.root)) { // we known branch start, common head is somewhere in its descendants line checkUp2Head.add(rb); } else { // dig deeper in the history, if necessary if (!rb.p1.isNull() && !localRepo.knownNode(rb.p1)) { toQuery.add(rb.p1); } if (!rb.p2.isNull() && !localRepo.knownNode(rb.p2)) { toQuery.add(rb.p2); } } } } // can't check nodes between checkUp2Head element and local heads, remote might have distinct descendants sequence for (RemoteBranch rb : checkUp2Head) { // rb.root is known locally List<Nodeid> remoteRevisions = remoteRepo.between(rb.head, rb.root); if (remoteRevisions.isEmpty()) { // head is immediate child resultCommon.add(rb.root); } else { // between gives result from head to root, I'd like to go in reverse direction Collections.reverse(remoteRevisions); Nodeid root = rb.root; while(!remoteRevisions.isEmpty()) { Nodeid n = remoteRevisions.remove(0); if (localRepo.knownNode(n)) { if (remoteRevisions.isEmpty()) { // this is the last known node before an unknown resultCommon.add(n); break; } if (remoteRevisions.size() == 1) { // there's only one left between known n and unknown head // this check is to save extra between query, not really essential Nodeid last = remoteRevisions.remove(0); resultCommon.add(localRepo.knownNode(last) ? last : n); break; } // might get handy for next between query, to narrow search down root = n; } else { remoteRevisions = remoteRepo.between(n, root); Collections.reverse(remoteRevisions); if (remoteRevisions.isEmpty()) { resultCommon.add(root); } } } } } // TODO ensure unique elements in the list return resultCommon; } // somewhat similar to Outgoing.findCommonWithRemote() public List<BranchChain> calculateMissingBranches() throws HgRemoteConnectionException { List<Nodeid> remoteHeads = remoteRepo.heads(); LinkedList<Nodeid> common = new LinkedList<Nodeid>(); // these remotes are known in local LinkedList<Nodeid> toQuery = new LinkedList<Nodeid>(); // these need further queries to find common for (Nodeid rh : remoteHeads) { if (localRepo.knownNode(rh)) { common.add(rh); } else { toQuery.add(rh); } } if (toQuery.isEmpty()) { return Collections.emptyList(); // no incoming changes } LinkedList<BranchChain> branches2load = new LinkedList<BranchChain>(); // return value // detailed comments are in Outgoing.findCommonWithRemote LinkedList<RemoteBranch> checkUp2Head = new LinkedList<RemoteBranch>(); // records relation between branch head and its parent branch, if any HashMap<Nodeid, BranchChain> head2chain = new HashMap<Nodeid, BranchChain>(); while (!toQuery.isEmpty()) { List<RemoteBranch> remoteBranches = remoteRepo.branches(toQuery); //head, root, first parent, second parent toQuery.clear(); while(!remoteBranches.isEmpty()) { RemoteBranch rb = remoteBranches.remove(0); BranchChain chainElement = head2chain.get(rb.head); if (chainElement == null) { chainElement = new BranchChain(rb.head); // record this unknown branch to download later branches2load.add(chainElement); // the only chance we'll need chainElement in the head2chain is when we know this branch's root head2chain.put(rb.head, chainElement); } if (localRepo.knownNode(rb.root)) { // we known branch start, common head is somewhere in its descendants line checkUp2Head.add(rb); } else { chainElement.branchRoot = rb.root; // dig deeper in the history, if necessary boolean hasP1 = !rb.p1.isNull(), hasP2 = !rb.p2.isNull(); if (hasP1 && !localRepo.knownNode(rb.p1)) { toQuery.add(rb.p1); // we might have seen parent node already, and recorded it as a branch chain // we shall reuse existing BC to get it completely initializer (head2chain map // on second put with the same key would leave first BC uninitialized. // It seems there's no reason to be affraid (XXX although shall double-check) // that BC's chain would get corrupt (its p1 and p2 fields assigned twice with different values) // as parents are always the same (and likely, BC that is common would be the last unknown) BranchChain bc = head2chain.get(rb.p1); if (bc == null) { head2chain.put(rb.p1, bc = new BranchChain(rb.p1)); } chainElement.p1 = bc; } if (hasP2 && !localRepo.knownNode(rb.p2)) { toQuery.add(rb.p2); BranchChain bc = head2chain.get(rb.p2); if (bc == null) { head2chain.put(rb.p2, bc = new BranchChain(rb.p2)); } chainElement.p2 = bc; } if (!hasP1 && !hasP2) { // special case, when we do incoming against blank repository, chainElement.branchRoot // is first unknown element (revision 0). We need to add another fake BranchChain // to fill the promise that terminal BranchChain has branchRoot that is known both locally and remotely BranchChain fake = new BranchChain(NULL); fake.branchRoot = NULL; chainElement.p1 = chainElement.p2 = fake; } } } } for (RemoteBranch rb : checkUp2Head) { Nodeid h = rb.head; Nodeid r = rb.root; int watchdog = 1000; assert head2chain.containsKey(h); BranchChain bc = head2chain.get(h); assert bc != null : h.toString(); // if we know branch root locally, there could be no parent branch chain elements. assert bc.p1 == null; assert bc.p2 == null; do { List<Nodeid> between = remoteRepo.between(h, r); if (between.isEmpty()) { bc.branchRoot = r; break; } else { Collections.reverse(between); for (Nodeid n : between) { if (localRepo.knownNode(n)) { r = n; } else { h = n; break; } } Nodeid lastInBetween = between.get(between.size() - 1); if (r.equals(lastInBetween)) { bc.branchRoot = r; break; } else if (h.equals(lastInBetween)) { // the only chance for current head pointer to point to the sequence tail // is when r is second from the between list end (iow, head,1,[2],4,8...,root) bc.branchRoot = r; break; } } } while(--watchdog > 0); if (watchdog == 0) { throw new HgInvalidStateException(String.format("Can't narrow down branch [%s, %s]", rb.head.shortNotation(), rb.root.shortNotation())); } } if (debug) { System.out.println("calculateMissingBranches:"); for (BranchChain bc : branches2load) { bc.dump(); } } return branches2load; } // root and head (and all between) are unknown for each chain element but last (terminal), which has known root (revision // known to be locally and at remote server // alternative would be to keep only unknown elements (so that promise of calculateMissingBranches would be 100% true), but that // seems to complicate the method, while being useful only for the case when we ask incoming for an empty repository (i.e. // where branch chain return all nodes, -1..tip. public static final class BranchChain { // when we construct a chain, we know head which is missing locally, hence init it right away. // as for root (branch unknown start), we might happen to have one locally, and need further digging to find out right branch start public final Nodeid branchHead; public Nodeid branchRoot; // either of these can be null, or both. // although RemoteBranch has either both parents null, or both non-null, when we construct a chain // we might encounter that we locally know one of branch's parent, hence in the chain corresponding field will be blank. public BranchChain p1; public BranchChain p2; public BranchChain(Nodeid head) { assert head != null; branchHead = head; } public boolean isTerminal() { return p1 == null && p2 == null; // either can be null, see comment above. Terminal is only when no way to descent } // true when this BranchChain is a branch that spans up to very start of the repository // Thus, the only common revision is NULL, recorded in a fake BranchChain object shared between p1 and p2 /*package-local*/ boolean isRepoStart() { return p1 == p2 && p1 != null && p1.branchHead == p1.branchRoot && p1.branchHead.isNull(); } @Override public String toString() { return String.format("BranchChain [%s, %s]", branchRoot, branchHead); } void dump() { System.out.println(toString()); internalDump(" "); } private void internalDump(String prefix) { if (p1 != null) { System.out.println(prefix + p1.toString()); } else if (p2 != null) { System.out.println(prefix + "NONE?!"); } if (p2 != null) { System.out.println(prefix + p2.toString()); } else if (p1 != null) { System.out.println(prefix + "NONE?!"); } prefix += " "; if (p1 != null) { p1.internalDump(prefix); } if (p2 != null) { p2.internalDump(prefix); } } } /** * @return list of nodeids from branchRoot to branchHead, inclusive. IOW, first element of the list is always root of the branch */ public List<Nodeid> completeBranch(final Nodeid branchRoot, final Nodeid branchHead) throws HgRemoteConnectionException { class DataEntry { public final Nodeid queryHead; public final int headIndex; public List<Nodeid> entries; public DataEntry(Nodeid head, int index, List<Nodeid> data) { queryHead = head; headIndex = index; entries = data; } }; List<Nodeid> initial = remoteRepo.between(branchHead, branchRoot); Nodeid[] result = new Nodeid[1 + (1 << initial.size())]; result[0] = branchHead; int rootIndex = -1; // index in the result, where to place branche's root. if (initial.isEmpty()) { rootIndex = 1; } else if (initial.size() == 1) { rootIndex = 2; } LinkedList<DataEntry> datas = new LinkedList<DataEntry>(); // DataEntry in datas has entries list filled with 'between' data, whereas // DataEntry in toQuery keeps only nodeid and its index, with entries to be initialized before // moving to datas. LinkedList<DataEntry> toQuery = new LinkedList<DataEntry>(); // datas.add(new DataEntry(branchHead, 0, initial)); int totalQueries = 1; HashSet<Nodeid> queried = new HashSet<Nodeid>(); while(!datas.isEmpty()) { // keep record of those planned to be queried next time we call between() // although may keep these in queried, if really don't want separate collection HashSet<Nodeid> scheduled = new HashSet<Nodeid>(); do { DataEntry de = datas.removeFirst(); // populate result with discovered elements between de.qiueryRoot and branch's head for (int i = 1, j = 0; j < de.entries.size(); i = i << 1, j++) { int idx = de.headIndex + i; result[idx] = de.entries.get(j); } // form next query entries from new unknown elements if (de.entries.size() > 1) { /* when entries has only one element, it means de.queryRoot was at head-2 position, and thus * no new information can be obtained. E.g. when it's 2, it might be case of [0..4] query with * [1,2] result, and we need one more query to get element 3. */ for (int i =1, j = 0; j < de.entries.size(); i = i<<1, j++) { int idx = de.headIndex + i; Nodeid x = de.entries.get(j); if (!queried.contains(x) && !scheduled.contains(x) && (rootIndex == -1 || rootIndex - de.headIndex > 1)) { /*queries for elements right before head is senseless, but unless we know head's index, do it anyway*/ toQuery.add(new DataEntry(x, idx, null)); scheduled.add(x); } } } } while (!datas.isEmpty()); if (!toQuery.isEmpty()) { totalQueries++; } // for each query, create an between request range, keep record Range->DataEntry to know range's start index LinkedList<HgRemoteRepository.Range> betweenBatch = new LinkedList<HgRemoteRepository.Range>(); HashMap<HgRemoteRepository.Range, DataEntry> rangeToEntry = new HashMap<HgRemoteRepository.Range, DataEntry>(); for (DataEntry de : toQuery) { queried.add(de.queryHead); HgRemoteRepository.Range r = new HgRemoteRepository.Range(branchRoot, de.queryHead); betweenBatch.add(r); rangeToEntry.put(r, de); } if (!betweenBatch.isEmpty()) { Map<Range, List<Nodeid>> between = remoteRepo.between(betweenBatch); for (Entry<Range, List<Nodeid>> e : between.entrySet()) { DataEntry de = rangeToEntry.get(e.getKey()); assert de != null; de.entries = e.getValue(); if (rootIndex == -1 && de.entries.size() == 1) { // returned sequence of length 1 means we used element from [head-2] as root int numberOfElementsExcludingRootAndHead = de.headIndex + 1; rootIndex = numberOfElementsExcludingRootAndHead + 1; if (debug) { System.out.printf("On query %d found out exact number of missing elements: %d\n", totalQueries, numberOfElementsExcludingRootAndHead); } } datas.add(de); // queue up to record result and construct further requests } betweenBatch.clear(); rangeToEntry.clear(); } toQuery.clear(); } if (rootIndex == -1) { throw new HgInvalidStateException("Shall not happen, provided between output is correct"); } result[rootIndex] = branchRoot; boolean resultOk = true; LinkedList<Nodeid> fromRootToHead = new LinkedList<Nodeid>(); IntVector missing = new IntVector(); for (int i = 0; i <= rootIndex; i++) { Nodeid n = result[i]; if (n == null) { missing.add(i); resultOk = false; } fromRootToHead.addFirst(n); // reverse order } if (debug) { System.out.println("Total queries:" + totalQueries); } if (!resultOk) { assert missing.size() > 0; // TODO post-1.0 perhaps, there's better alternative than HgInvalidStateException, e.g. HgDataFormatException? throw new HgInvalidStateException(String.format("Missing elements with indexes: %s", Arrays.toString(missing.toArray()))); } return fromRootToHead; } /** * returns in order from branch root to head * for a non-empty BranchChain, shall return modifiable list */ public List<Nodeid> visitBranches(BranchChain bc) throws HgRemoteConnectionException { if (bc == null) { return Collections.emptyList(); } List<Nodeid> mine = completeBranch(bc.branchRoot, bc.branchHead); if (bc.isTerminal() || bc.isRepoStart()) { return mine; } List<Nodeid> parentBranch1 = visitBranches(bc.p1); List<Nodeid> parentBranch2 = visitBranches(bc.p2); // merge LinkedList<Nodeid> merged = new LinkedList<Nodeid>(); ListIterator<Nodeid> i1 = parentBranch1.listIterator(), i2 = parentBranch2.listIterator(); while (i1.hasNext() && i2.hasNext()) { Nodeid n1 = i1.next(); Nodeid n2 = i2.next(); if (n1.equals(n2)) { merged.addLast(n1); } else { // first different => add both, and continue adding both tails sequentially merged.add(n2); merged.add(n1); break; } } // copy rest of second parent branch while (i2.hasNext()) { merged.add(i2.next()); } // copy rest of first parent branch while (i1.hasNext()) { merged.add(i1.next()); } // ArrayList<Nodeid> rv = new ArrayList<Nodeid>(mine.size() + merged.size()); rv.addAll(merged); rv.addAll(mine); return rv; } public void collectKnownRoots(BranchChain bc, Set<Nodeid> result) { if (bc == null) { return; } if (bc.isTerminal()) { result.add(bc.branchRoot); return; } if (bc.isRepoStart()) { return; } collectKnownRoots(bc.p1, result); collectKnownRoots(bc.p2, result); } }