Mercurial > hg4j
view cmdline/org/tmatesoft/hg/console/Incoming.java @ 178:62665d8f0686
Complete logic to discover all branches missing locally. Most of wire protocol in HgRemoteRepository
author | Artem Tikhomirov <tikhomirov.artem@gmail.com> |
---|---|
date | Wed, 06 Apr 2011 01:34:16 +0200 |
parents | e10225daface |
children | cd3371670f0b |
line wrap: on
line source
/* * Copyright (c) 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.console; import static org.tmatesoft.hg.core.Nodeid.NULL; import java.io.File; import java.net.URL; import java.util.ArrayList; import java.util.Arrays; import java.util.Collections; import java.util.Comparator; import java.util.HashMap; import java.util.HashSet; import java.util.Iterator; import java.util.LinkedHashMap; import java.util.LinkedList; import java.util.List; import java.util.ListIterator; import java.util.Map; import java.util.Map.Entry; import org.tmatesoft.hg.core.HgBadStateException; import org.tmatesoft.hg.core.HgException; import org.tmatesoft.hg.core.Nodeid; import org.tmatesoft.hg.internal.ConfigFile; import org.tmatesoft.hg.internal.Internals; import org.tmatesoft.hg.repo.HgChangelog; import org.tmatesoft.hg.repo.HgLookup; 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.HgRepository; /** * WORK IN PROGRESS, DO NOT USE * hg incoming counterpart * * @author Artem Tikhomirov * @author TMate Software Ltd. */ public class Incoming { public static void main(String[] args) throws Exception { if (Boolean.FALSE.booleanValue()) { new SequenceConstructor().test(); return; } Options cmdLineOpts = Options.parse(args); HgRepository hgRepo = cmdLineOpts.findRepository(); if (hgRepo.isInvalid()) { System.err.printf("Can't find repository in: %s\n", hgRepo.getLocation()); return; } String key = "svnkit"; ConfigFile cfg = new Internals().newConfigFile(); cfg.addLocation(new File(System.getProperty("user.home"), ".hgrc")); String server = cfg.getSection("paths").get(key); if (server == null) { throw new HgException(String.format("Can't find server %s specification in the config", key)); } HgRemoteRepository hgRemote = new HgLookup().detect(new URL(server)); // // in fact, all we need from changelog is set of all nodeids. However, since ParentWalker reuses same Nodeids, it's not too expensive // to reuse it here, XXX although later this may need to be refactored final HgChangelog.ParentWalker pw = hgRepo.getChangelog().new ParentWalker(); pw.init(); // List<BranchChain> missingBranches0 = calculateMissingBranches(pw, hgRemote); for (BranchChain bc : missingBranches0) { bc.dump(); List<Nodeid> missing = visitBranches(hgRemote, bc); // Collections.reverse(missing); // useful to test output, from newer to older for (Nodeid n : missing) { if (pw.knownNode(n)) { System.out.println("Erroneous to fetch:" + n); } else { System.out.println(n); } } System.out.println("Branch done"); } } private static 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; } @Override public String toString() { return String.format("BranchChain [%s, %s]", branchRoot, branchHead); } void dump() { System.out.println(toString()); internalDump(" "); } void internalDump(String prefix) { if (p1 != null) { System.out.println(prefix + p1.toString()); } if (p2 != null) { System.out.println(prefix + p2.toString()); } prefix += " "; if (p1 != null) { p1.internalDump(prefix); } if (p2 != null) { p2.internalDump(prefix); } } } private static List<Nodeid> visitBranches(HgRemoteRepository hgRemote, BranchChain bc) throws HgException { if (bc == null) { return Collections.emptyList(); } List<Nodeid> mine = completeBranch(hgRemote, bc.branchRoot, bc.branchHead); if (bc.isTerminal()) { return mine; } List<Nodeid> parentBranch1 = visitBranches(hgRemote, bc.p1); List<Nodeid> parentBranch2 = visitBranches(hgRemote, 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; } // somewhat similar to Outgoing.findCommonWithRemote() private static List<BranchChain> calculateMissingBranches(HgChangelog.ParentWalker pwLocal, HgRemoteRepository hgRemote) throws HgException { List<Nodeid> remoteHeads = hgRemote.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 (pwLocal.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, Incoming.BranchChain>(); while (!toQuery.isEmpty()) { List<RemoteBranch> remoteBranches = hgRemote.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); } if (pwLocal.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 if (!NULL.equals(rb.p1) && !pwLocal.knownNode(rb.p1)) { toQuery.add(rb.p1); head2chain.put(rb.p1, chainElement.p1 = new BranchChain(rb.p1)); } if (!NULL.equals(rb.p2) && !pwLocal.knownNode(rb.p2)) { toQuery.add(rb.p2); head2chain.put(rb.p2, chainElement.p2 = new BranchChain(rb.p2)); } } } } for (RemoteBranch rb : checkUp2Head) { Nodeid h = rb.head; Nodeid r = rb.root; int watchdog = 1000; BranchChain bc = head2chain.get(h); assert bc != null; // 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 = hgRemote.between(h, r); if (between.isEmpty()) { bc.branchRoot = r; break; } else { Collections.reverse(between); for (Nodeid n : between) { if (pwLocal.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 HgBadStateException(String.format("Can't narrow down branch [%s, %s]", rb.head.shortNotation(), rb.root.shortNotation())); } } return branches2load; } /** * @return list of nodeids from branchRoot to branchHead, inclusive. IOW, first element of the list is always root of the branch */ private static List<Nodeid> completeBranch(HgRemoteRepository hgRemote, final Nodeid branchRoot, final Nodeid branchHead) throws HgException { 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 = hgRemote.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 = hgRemote.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; 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 HgBadStateException("Shall not happen, provided between output is correct"); // FIXME } result[rootIndex] = branchRoot; boolean resultOk = true; LinkedList<Nodeid> fromRootToHead = new LinkedList<Nodeid>(); for (int i = 0; i <= rootIndex; i++) { Nodeid n = result[i]; if (n == null) { System.out.printf("ERROR: element %d wasn't found\n",i); resultOk = false; } fromRootToHead.addFirst(n); // reverse order } System.out.println("Total queries:" + totalQueries); if (!resultOk) { throw new HgBadStateException("See console for details"); // FIXME } return fromRootToHead; } private static class SequenceConstructor { private int[] between(int root, int head) { if (head <= (root+1)) { return new int[0]; } System.out.printf("[%d, %d]\t\t", root, head); int size = 1 + (int) Math.floor(Math.log(head-root - 1) / Math.log(2)); int[] rv = new int[size]; for (int v = 1, i = 0; i < rv.length; i++) { rv[i] = root + v; v = v << 1; } System.out.println(Arrays.toString(rv)); return rv; } public void test() { int root = 0, head = 126; int[] data = between(root, head); // max number of elements to recover is 2**data.length-1, when head is exactly // 2**data.length element of the branch. // In such case, total number of elements in the branch (including head and root, would be 2**data.length+1 int[] finalSequence = new int[1 + (1 << data.length >>> 5)]; // div 32 - total bits to integers, +1 for possible modulus int exactNumberOfElements = -1; // exact number of meaningful bits in finalSequence LinkedHashMap<Integer, int[]> datas = new LinkedHashMap<Integer, int[]>(); datas.put(root, data); int totalQueries = 1; HashSet<Integer> queried = new HashSet<Integer>(); int[] checkSequence = null; while(!datas.isEmpty()) { LinkedList<int[]> toQuery = new LinkedList<int[]>(); do { Iterator<Entry<Integer, int[]>> it = datas.entrySet().iterator(); Entry<Integer, int[]> next = it.next(); int r = next.getKey(); data = next.getValue(); it.remove(); populate(r, head, data, finalSequence); if (checkSequence != null) { boolean match = true; // System.out.println("Try to match:"); for (int i = 0; i < checkSequence.length; i++) { // System.out.println(i); // System.out.println("control:" + toBinaryString(checkSequence[i], ' ')); // System.out.println("present:" + toBinaryString(finalSequence[i], ' ')); if (checkSequence[i] != finalSequence[i]) { match = false; } else { match &= true; } } System.out.println(match ? "Match, on query:" + totalQueries : "Didn't match"); } if (data.length > 1) { /*queries for elements next to head is senseless, hence data.length check above and head-x below*/ for (int x : data) { if (!queried.contains(x) && head - x > 1) { toQuery.add(new int[] {x, head}); } } } } while (!datas.isEmpty()) ; if (!toQuery.isEmpty()) { System.out.println(); totalQueries++; } Collections.sort(toQuery, new Comparator<int[]>() { public int compare(int[] o1, int[] o2) { return o1[0] < o2[0] ? -1 : (o1[0] == o2[0] ? 0 : 1); } }); for (int[] x : toQuery) { if (!queried.contains(x[0])) { queried.add(x[0]); data = between(x[0], x[1]); if (exactNumberOfElements == -1 && data.length == 1) { exactNumberOfElements = x[0] + 1; System.out.printf("On query %d found out exact number of missing elements: %d\n", totalQueries, exactNumberOfElements); // get a bit sequence of exactNumberOfElements, 0111..110 // to 'and' it with finalSequence later int totalInts = (exactNumberOfElements + 2 /*heading and tailing zero bits*/) >>> 5; int trailingBits = (exactNumberOfElements + 2) & 0x1f; if (trailingBits != 0) { totalInts++; } checkSequence = new int[totalInts]; Arrays.fill(checkSequence, 0xffffffff); checkSequence[0] &= 0x7FFFFFFF; if (trailingBits == 0) { checkSequence[totalInts-1] &= 0xFFFFFFFE; } else if (trailingBits == 1) { checkSequence[totalInts-1] = 0; } else { // trailingBits include heading and trailing zero bits int mask = 0x80000000 >> trailingBits-2; // with sign! checkSequence[totalInts - 1] &= mask; } for (int e : checkSequence) { System.out.print(toBinaryString(e, ' ')); } System.out.println(); } datas.put(x[0], data); } } } System.out.println("Total queries:" + totalQueries); for (int x : finalSequence) { System.out.print(toBinaryString(x, ' ')); } } private void populate(int root, int head, int[] data, int[] finalSequence) { for (int i = 1, x = 0; root+i < head; i = i << 1, x++) { int value = data[x]; int value_check = root+i; if (value != value_check) { throw new IllegalStateException(); } int wordIx = (root + i) >>> 5; int bitIx = (root + i) & 0x1f; finalSequence[wordIx] |= 1 << (31-bitIx); } } private static String toBinaryString(int x, char byteSeparator) { StringBuilder sb = new StringBuilder(4*8+4); sb.append(toBinaryString((byte) (x >>> 24))); sb.append(byteSeparator); sb.append(toBinaryString((byte) ((x & 0x00ff0000) >>> 16))); sb.append(byteSeparator); sb.append(toBinaryString((byte) ((x & 0x00ff00) >>> 8))); sb.append(byteSeparator); sb.append(toBinaryString((byte) (x & 0x00ff))); sb.append(byteSeparator); return sb.toString(); } private static String toBinaryString(byte b) { final String nibbles = "0000000100100011010001010110011110001001101010111100110111101111"; assert nibbles.length() == 16*4; int x1 = (b >>> 4) & 0x0f, x2 = b & 0x0f; x1 *= 4; x2 *= 4; // 4 characters per nibble return nibbles.substring(x1, x1+4).concat(nibbles.substring(x2, x2+4)); } } }