Package org.eclipse.jgit.lib

Source Code of org.eclipse.jgit.lib.Repository

/*
* Copyright (C) 2007, Dave Watson <dwatson@mimvista.com>
* Copyright (C) 2008-2010, Google Inc.
* Copyright (C) 2006-2010, Robin Rosenberg <robin.rosenberg@dewire.com>
* Copyright (C) 2006-2008, Shawn O. Pearce <spearce@spearce.org>
* and other copyright owners as documented in the project's IP log.
*
* This program and the accompanying materials are made available
* under the terms of the Eclipse Distribution License v1.0 which
* accompanies this distribution, is reproduced below, and is
* available at http://www.eclipse.org/org/documents/edl-v10.php
*
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or
* without modification, are permitted provided that the following
* conditions are met:
*
* - Redistributions of source code must retain the above copyright
*   notice, this list of conditions and the following disclaimer.
*
* - Redistributions in binary form must reproduce the above
*   copyright notice, this list of conditions and the following
*   disclaimer in the documentation and/or other materials provided
*   with the distribution.
*
* - Neither the name of the Eclipse Foundation, Inc. nor the
*   names of its contributors may be used to endorse or promote
*   products derived from this software without specific prior
*   written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND
* CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES,
* INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
* STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
* ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/

package org.eclipse.jgit.lib;

import java.io.BufferedOutputStream;
import java.io.File;
import java.io.FileNotFoundException;
import java.io.FileOutputStream;
import java.io.IOException;
import java.text.MessageFormat;
import java.util.Collection;
import java.util.Collections;
import java.util.HashMap;
import java.util.HashSet;
import java.util.LinkedList;
import java.util.List;
import java.util.Map;
import java.util.Set;
import java.util.concurrent.atomic.AtomicInteger;

import org.eclipse.jgit.JGitText;
import org.eclipse.jgit.dircache.DirCache;
import org.eclipse.jgit.errors.AmbiguousObjectException;
import org.eclipse.jgit.errors.CorruptObjectException;
import org.eclipse.jgit.errors.IncorrectObjectTypeException;
import org.eclipse.jgit.errors.MissingObjectException;
import org.eclipse.jgit.errors.NoWorkTreeException;
import org.eclipse.jgit.errors.RevisionSyntaxException;
import org.eclipse.jgit.events.IndexChangedEvent;
import org.eclipse.jgit.events.IndexChangedListener;
import org.eclipse.jgit.events.ListenerList;
import org.eclipse.jgit.events.RepositoryEvent;
import org.eclipse.jgit.revwalk.RevBlob;
import org.eclipse.jgit.revwalk.RevCommit;
import org.eclipse.jgit.revwalk.RevObject;
import org.eclipse.jgit.revwalk.RevTree;
import org.eclipse.jgit.revwalk.RevWalk;
import org.eclipse.jgit.storage.file.ReflogEntry;
import org.eclipse.jgit.storage.file.ReflogReader;
import org.eclipse.jgit.treewalk.TreeWalk;
import org.eclipse.jgit.util.FS;
import org.eclipse.jgit.util.FileUtils;
import org.eclipse.jgit.util.IO;
import org.eclipse.jgit.util.RawParseUtils;
import org.eclipse.jgit.util.io.SafeBufferedOutputStream;

/**
* Represents a Git repository.
* <p>
* A repository holds all objects and refs used for managing source code (could
* be any type of file, but source code is what SCM's are typically used for).
* <p>
* This class is thread-safe.
*/
public abstract class Repository {
  private static final ListenerList globalListeners = new ListenerList();

  /** @return the global listener list observing all events in this JVM. */
  public static ListenerList getGlobalListenerList() {
    return globalListeners;
  }

  private final AtomicInteger useCnt = new AtomicInteger(1);

  /** Metadata directory holding the repository's critical files. */
  private final File gitDir;

  /** File abstraction used to resolve paths. */
  private final FS fs;

  private final ListenerList myListeners = new ListenerList();

  /** If not bare, the top level directory of the working files. */
  private final File workTree;

  /** If not bare, the index file caching the working file states. */
  private final File indexFile;

  /**
   * Initialize a new repository instance.
   *
   * @param options
   *            options to configure the repository.
   */
  protected Repository(final BaseRepositoryBuilder options) {
    gitDir = options.getGitDir();
    fs = options.getFS();
    workTree = options.getWorkTree();
    indexFile = options.getIndexFile();
  }

  /** @return listeners observing only events on this repository. */
  public ListenerList getListenerList() {
    return myListeners;
  }

  /**
   * Fire an event to all registered listeners.
   * <p>
   * The source repository of the event is automatically set to this
   * repository, before the event is delivered to any listeners.
   *
   * @param event
   *            the event to deliver.
   */
  public void fireEvent(RepositoryEvent<?> event) {
    event.setRepository(this);
    myListeners.dispatch(event);
    globalListeners.dispatch(event);
  }

  /**
   * Create a new Git repository.
   * <p>
   * Repository with working tree is created using this method. This method is
   * the same as {@code create(false)}.
   *
   * @throws IOException
   * @see #create(boolean)
   */
  public void create() throws IOException {
    create(false);
  }

  /**
   * Create a new Git repository initializing the necessary files and
   * directories.
   *
   * @param bare
   *            if true, a bare repository (a repository without a working
   *            directory) is created.
   * @throws IOException
   *             in case of IO problem
   */
  public abstract void create(boolean bare) throws IOException;

  /** @return local metadata directory; null if repository isn't local. */
  public File getDirectory() {
    return gitDir;
  }

  /**
   * @return the object database which stores this repository's data.
   */
  public abstract ObjectDatabase getObjectDatabase();

  /** @return a new inserter to create objects in {@link #getObjectDatabase()} */
  public ObjectInserter newObjectInserter() {
    return getObjectDatabase().newInserter();
  }

  /** @return a new reader to read objects from {@link #getObjectDatabase()} */
  public ObjectReader newObjectReader() {
    return getObjectDatabase().newReader();
  }

  /** @return the reference database which stores the reference namespace. */
  public abstract RefDatabase getRefDatabase();

  /**
   * @return the configuration of this repository
   */
  public abstract StoredConfig getConfig();

  /**
   * @return the used file system abstraction
   */
  public FS getFS() {
    return fs;
  }

  /**
   * @param objectId
   * @return true if the specified object is stored in this repo or any of the
   *         known shared repositories.
   */
  public boolean hasObject(AnyObjectId objectId) {
    try {
      return getObjectDatabase().has(objectId);
    } catch (IOException e) {
      // Legacy API, assume error means "no"
      return false;
    }
  }

  /**
   * Open an object from this repository.
   * <p>
   * This is a one-shot call interface which may be faster than allocating a
   * {@link #newObjectReader()} to perform the lookup.
   *
   * @param objectId
   *            identity of the object to open.
   * @return a {@link ObjectLoader} for accessing the object.
   * @throws MissingObjectException
   *             the object does not exist.
   * @throws IOException
   *             the object store cannot be accessed.
   */
  public ObjectLoader open(final AnyObjectId objectId)
      throws MissingObjectException, IOException {
    return getObjectDatabase().open(objectId);
  }

  /**
   * Open an object from this repository.
   * <p>
   * This is a one-shot call interface which may be faster than allocating a
   * {@link #newObjectReader()} to perform the lookup.
   *
   * @param objectId
   *            identity of the object to open.
   * @param typeHint
   *            hint about the type of object being requested;
   *            {@link ObjectReader#OBJ_ANY} if the object type is not known,
   *            or does not matter to the caller.
   * @return a {@link ObjectLoader} for accessing the object.
   * @throws MissingObjectException
   *             the object does not exist.
   * @throws IncorrectObjectTypeException
   *             typeHint was not OBJ_ANY, and the object's actual type does
   *             not match typeHint.
   * @throws IOException
   *             the object store cannot be accessed.
   */
  public ObjectLoader open(AnyObjectId objectId, int typeHint)
      throws MissingObjectException, IncorrectObjectTypeException,
      IOException {
    return getObjectDatabase().open(objectId, typeHint);
  }

  /**
   * Create a command to update, create or delete a ref in this repository.
   *
   * @param ref
   *            name of the ref the caller wants to modify.
   * @return an update command. The caller must finish populating this command
   *         and then invoke one of the update methods to actually make a
   *         change.
   * @throws IOException
   *             a symbolic ref was passed in and could not be resolved back
   *             to the base ref, as the symbolic ref could not be read.
   */
  public RefUpdate updateRef(final String ref) throws IOException {
    return updateRef(ref, false);
  }

  /**
   * Create a command to update, create or delete a ref in this repository.
   *
   * @param ref
   *            name of the ref the caller wants to modify.
   * @param detach
   *            true to create a detached head
   * @return an update command. The caller must finish populating this command
   *         and then invoke one of the update methods to actually make a
   *         change.
   * @throws IOException
   *             a symbolic ref was passed in and could not be resolved back
   *             to the base ref, as the symbolic ref could not be read.
   */
  public RefUpdate updateRef(final String ref, final boolean detach) throws IOException {
    return getRefDatabase().newUpdate(ref, detach);
  }

  /**
   * Create a command to rename a ref in this repository
   *
   * @param fromRef
   *            name of ref to rename from
   * @param toRef
   *            name of ref to rename to
   * @return an update command that knows how to rename a branch to another.
   * @throws IOException
   *             the rename could not be performed.
   *
   */
  public RefRename renameRef(final String fromRef, final String toRef) throws IOException {
    return getRefDatabase().newRename(fromRef, toRef);
  }

  /**
   * Parse a git revision string and return an object id.
   *
   * Combinations of these operators are supported:
   * <ul>
   * <li><b>HEAD</b>, <b>MERGE_HEAD</b>, <b>FETCH_HEAD</b></li>
   * <li><b>SHA-1</b>: a complete or abbreviated SHA-1</li>
   * <li><b>refs/...</b>: a complete reference name</li>
   * <li><b>short-name</b>: a short reference name under {@code refs/heads},
   * {@code refs/tags}, or {@code refs/remotes} namespace</li>
   * <li><b>tag-NN-gABBREV</b>: output from describe, parsed by treating
   * {@code ABBREV} as an abbreviated SHA-1.</li>
   * <li><i>id</i><b>^</b>: first parent of commit <i>id</i>, this is the same
   * as {@code id^1}</li>
   * <li><i>id</i><b>^0</b>: ensure <i>id</i> is a commit</li>
   * <li><i>id</i><b>^n</b>: n-th parent of commit <i>id</i></li>
   * <li><i>id</i><b>~n</b>: n-th historical ancestor of <i>id</i>, by first
   * parent. {@code id~3} is equivalent to {@code id^1^1^1} or {@code id^^^}.</li>
   * <li><i>id</i><b>:path</b>: Lookup path under tree named by <i>id</i></li>
   * <li><i>id</i><b>^{commit}</b>: ensure <i>id</i> is a commit</li>
   * <li><i>id</i><b>^{tree}</b>: ensure <i>id</i> is a tree</li>
   * <li><i>id</i><b>^{tag}</b>: ensure <i>id</i> is a tag</li>
   * <li><i>id</i><b>^{blob}</b>: ensure <i>id</i> is a blob</li>
   * </ul>
   *
   * <p>
   * The following operators are specified by Git conventions, but are not
   * supported by this method:
   * <ul>
   * <li><b>ref@{n}</b>: n-th version of ref as given by its reflog</li>
   * <li><b>ref@{time}</b>: value of ref at the designated time</li>
   * </ul>
   *
   * @param revstr
   *            A git object references expression
   * @return an ObjectId or null if revstr can't be resolved to any ObjectId
   * @throws AmbiguousObjectException
   *             {@code revstr} contains an abbreviated ObjectId and this
   *             repository contains more than one object which match to the
   *             input abbreviation.
   * @throws IncorrectObjectTypeException
   *             the id parsed does not meet the type required to finish
   *             applying the operators in the expression.
   * @throws RevisionSyntaxException
   *             the expression is not supported by this implementation, or
   *             does not meet the standard syntax.
   * @throws IOException
   *             on serious errors
   */
  public ObjectId resolve(final String revstr)
      throws AmbiguousObjectException, IOException {
    RevWalk rw = new RevWalk(this);
    try {
      return resolve(rw, revstr);
    } finally {
      rw.release();
    }
  }

  private ObjectId resolve(final RevWalk rw, final String revstr) throws IOException {
    char[] rev = revstr.toCharArray();
    RevObject ref = null;
    for (int i = 0; i < rev.length; ++i) {
      switch (rev[i]) {
      case '^':
        if (ref == null) {
          ref = parseSimple(rw, new String(rev, 0, i));
          if (ref == null)
            return null;
        }
        if (i + 1 < rev.length) {
          switch (rev[i + 1]) {
          case '0':
          case '1':
          case '2':
          case '3':
          case '4':
          case '5':
          case '6':
          case '7':
          case '8':
          case '9':
            int j;
            ref = rw.parseCommit(ref);
            for (j = i + 1; j < rev.length; ++j) {
              if (!Character.isDigit(rev[j]))
                break;
            }
            String parentnum = new String(rev, i + 1, j - i - 1);
            int pnum;
            try {
              pnum = Integer.parseInt(parentnum);
            } catch (NumberFormatException e) {
              throw new RevisionSyntaxException(
                  JGitText.get().invalidCommitParentNumber,
                  revstr);
            }
            if (pnum != 0) {
              RevCommit commit = (RevCommit) ref;
              if (pnum > commit.getParentCount())
                ref = null;
              else
                ref = commit.getParent(pnum - 1);
            }
            i = j - 1;
            break;
          case '{':
            int k;
            String item = null;
            for (k = i + 2; k < rev.length; ++k) {
              if (rev[k] == '}') {
                item = new String(rev, i + 2, k - i - 2);
                break;
              }
            }
            i = k;
            if (item != null)
              if (item.equals("tree")) {
                ref = rw.parseTree(ref);
              } else if (item.equals("commit")) {
                ref = rw.parseCommit(ref);
              } else if (item.equals("blob")) {
                ref = rw.peel(ref);
                if (!(ref instanceof RevBlob))
                  throw new IncorrectObjectTypeException(ref,
                      Constants.TYPE_BLOB);
              } else if (item.equals("")) {
                ref = rw.peel(ref);
              } else
                throw new RevisionSyntaxException(revstr);
            else
              throw new RevisionSyntaxException(revstr);
            break;
          default:
            ref = rw.parseAny(ref);
            if (ref instanceof RevCommit) {
              RevCommit commit = ((RevCommit) ref);
              if (commit.getParentCount() == 0)
                ref = null;
              else
                ref = commit.getParent(0);
            } else
              throw new IncorrectObjectTypeException(ref,
                  Constants.TYPE_COMMIT);

          }
        } else {
          ref = rw.peel(ref);
          if (ref instanceof RevCommit) {
            RevCommit commit = ((RevCommit) ref);
            if (commit.getParentCount() == 0)
              ref = null;
            else
              ref = commit.getParent(0);
          } else
            throw new IncorrectObjectTypeException(ref,
                Constants.TYPE_COMMIT);
        }
        break;
      case '~':
        if (ref == null) {
          ref = parseSimple(rw, new String(rev, 0, i));
          if (ref == null)
            return null;
        }
        ref = rw.peel(ref);
        if (!(ref instanceof RevCommit))
          throw new IncorrectObjectTypeException(ref,
              Constants.TYPE_COMMIT);
        int l;
        for (l = i + 1; l < rev.length; ++l) {
          if (!Character.isDigit(rev[l]))
            break;
        }
        int dist;
        if (l - i > 1) {
          String distnum = new String(rev, i + 1, l - i - 1);
          try {
            dist = Integer.parseInt(distnum);
          } catch (NumberFormatException e) {
            throw new RevisionSyntaxException(
                JGitText.get().invalidAncestryLength, revstr);
          }
        } else
          dist = 1;
        while (dist > 0) {
          RevCommit commit = (RevCommit) ref;
          if (commit.getParentCount() == 0) {
            ref = null;
            break;
          }
          commit = commit.getParent(0);
          rw.parseHeaders(commit);
          ref = commit;
          --dist;
        }
        i = l - 1;
        break;
      case '@':
        int m;
        String time = null;
        for (m = i + 2; m < rev.length; ++m) {
          if (rev[m] == '}') {
            time = new String(rev, i + 2, m - i - 2);
            break;
          }
        }
        if (time != null) {
          String refName = new String(rev, 0, i);
          Ref resolved = getRefDatabase().getRef(refName);
          if (resolved == null)
            return null;
          ref = resolveReflog(rw, resolved, time);
          i = m;
        } else
          i = m - 1;
        break;
      case ':': {
        RevTree tree;
        if (ref == null) {
          // We might not yet have parsed the left hand side.
          ObjectId id;
          try {
            if (i == 0)
              id = resolve(rw, Constants.HEAD);
            else
              id = resolve(rw, new String(rev, 0, i));
          } catch (RevisionSyntaxException badSyntax) {
            throw new RevisionSyntaxException(revstr);
          }
          if (id == null)
            return null;
          tree = rw.parseTree(id);
        } else {
          tree = rw.parseTree(ref);
        }

        if (i == rev.length - 1)
          return tree.copy();

        TreeWalk tw = TreeWalk.forPath(rw.getObjectReader(),
            new String(rev, i + 1, rev.length - i - 1), tree);
        return tw != null ? tw.getObjectId(0) : null;
      }

      default:
        if (ref != null)
          throw new RevisionSyntaxException(revstr);
      }
    }
    return ref != null ? ref.copy() : resolveSimple(revstr);
  }

  private static boolean isHex(char c) {
    return ('0' <= c && c <= '9') //
        || ('a' <= c && c <= 'f') //
        || ('A' <= c && c <= 'F');
  }

  private static boolean isAllHex(String str, int ptr) {
    while (ptr < str.length()) {
      if (!isHex(str.charAt(ptr++)))
        return false;
    }
    return true;
  }

  private RevObject parseSimple(RevWalk rw, String revstr) throws IOException {
    ObjectId id = resolveSimple(revstr);
    return id != null ? rw.parseAny(id) : null;
  }

  private ObjectId resolveSimple(final String revstr) throws IOException {
    if (ObjectId.isId(revstr))
      return ObjectId.fromString(revstr);

    Ref r = getRefDatabase().getRef(revstr);
    if (r != null)
      return r.getObjectId();

    if (AbbreviatedObjectId.isId(revstr))
      return resolveAbbreviation(revstr);

    int dashg = revstr.indexOf("-g");
    if ((dashg + 5) < revstr.length() && 0 <= dashg
        && isHex(revstr.charAt(dashg + 2))
        && isHex(revstr.charAt(dashg + 3))
        && isAllHex(revstr, dashg + 4)) {
      // Possibly output from git describe?
      String s = revstr.substring(dashg + 2);
      if (AbbreviatedObjectId.isId(s))
        return resolveAbbreviation(s);
    }

    return null;
  }

  private RevCommit resolveReflog(RevWalk rw, Ref ref, String time)
      throws IOException {
    int number;
    try {
      number = Integer.parseInt(time);
    } catch (NumberFormatException nfe) {
      throw new RevisionSyntaxException(MessageFormat.format(
          JGitText.get().invalidReflogRevision, time));
    }
    if (number < 0)
      throw new RevisionSyntaxException(MessageFormat.format(
          JGitText.get().invalidReflogRevision, time));

    ReflogReader reader = new ReflogReader(this, ref.getName());
    ReflogEntry entry = reader.getReverseEntry(number);
    if (entry == null)
      throw new RevisionSyntaxException(MessageFormat.format(
          JGitText.get().reflogEntryNotFound,
          Integer.valueOf(number), ref.getName()));

    return rw.parseCommit(entry.getNewId());
  }

  private ObjectId resolveAbbreviation(final String revstr) throws IOException,
      AmbiguousObjectException {
    AbbreviatedObjectId id = AbbreviatedObjectId.fromString(revstr);
    ObjectReader reader = newObjectReader();
    try {
      Collection<ObjectId> matches = reader.resolve(id);
      if (matches.size() == 0)
        return null;
      else if (matches.size() == 1)
        return matches.iterator().next();
      else
        throw new AmbiguousObjectException(id, matches);
    } finally {
      reader.release();
    }
  }

  /** Increment the use counter by one, requiring a matched {@link #close()}. */
  public void incrementOpen() {
    useCnt.incrementAndGet();
  }

  /** Decrement the use count, and maybe close resources. */
  public void close() {
    if (useCnt.decrementAndGet() == 0) {
      doClose();
    }
  }

  /**
   * Invoked when the use count drops to zero during {@link #close()}.
   * <p>
   * The default implementation closes the object and ref databases.
   */
  protected void doClose() {
    getObjectDatabase().close();
    getRefDatabase().close();
  }

  public String toString() {
    String desc;
    if (getDirectory() != null)
      desc = getDirectory().getPath();
    else
      desc = getClass().getSimpleName() + "-"
          + System.identityHashCode(this);
    return "Repository[" + desc + "]";
  }

  /**
   * Get the name of the reference that {@code HEAD} points to.
   * <p>
   * This is essentially the same as doing:
   *
   * <pre>
   * return getRef(Constants.HEAD).getTarget().getName()
   * </pre>
   *
   * Except when HEAD is detached, in which case this method returns the
   * current ObjectId in hexadecimal string format.
   *
   * @return name of current branch (for example {@code refs/heads/master}) or
   *         an ObjectId in hex format if the current branch is detached.
   * @throws IOException
   */
  public String getFullBranch() throws IOException {
    Ref head = getRef(Constants.HEAD);
    if (head == null)
      return null;
    if (head.isSymbolic())
      return head.getTarget().getName();
    if (head.getObjectId() != null)
      return head.getObjectId().name();
    return null;
  }

  /**
   * Get the short name of the current branch that {@code HEAD} points to.
   * <p>
   * This is essentially the same as {@link #getFullBranch()}, except the
   * leading prefix {@code refs/heads/} is removed from the reference before
   * it is returned to the caller.
   *
   * @return name of current branch (for example {@code master}), or an
   *         ObjectId in hex format if the current branch is detached.
   * @throws IOException
   */
  public String getBranch() throws IOException {
    String name = getFullBranch();
    if (name != null)
      return shortenRefName(name);
    return name;
  }

  /**
   * Objects known to exist but not expressed by {@link #getAllRefs()}.
   * <p>
   * When a repository borrows objects from another repository, it can
   * advertise that it safely has that other repository's references, without
   * exposing any other details about the other repository.  This may help
   * a client trying to push changes avoid pushing more than it needs to.
   *
   * @return unmodifiable collection of other known objects.
   */
  public Set<ObjectId> getAdditionalHaves() {
    return Collections.emptySet();
  }

  /**
   * Get a ref by name.
   *
   * @param name
   *            the name of the ref to lookup. May be a short-hand form, e.g.
   *            "master" which is is automatically expanded to
   *            "refs/heads/master" if "refs/heads/master" already exists.
   * @return the Ref with the given name, or null if it does not exist
   * @throws IOException
   */
  public Ref getRef(final String name) throws IOException {
    return getRefDatabase().getRef(name);
  }

  /**
   * @return mutable map of all known refs (heads, tags, remotes).
   */
  public Map<String, Ref> getAllRefs() {
    try {
      return getRefDatabase().getRefs(RefDatabase.ALL);
    } catch (IOException e) {
      return new HashMap<String, Ref>();
    }
  }

  /**
   * @return mutable map of all tags; key is short tag name ("v1.0") and value
   *         of the entry contains the ref with the full tag name
   *         ("refs/tags/v1.0").
   */
  public Map<String, Ref> getTags() {
    try {
      return getRefDatabase().getRefs(Constants.R_TAGS);
    } catch (IOException e) {
      return new HashMap<String, Ref>();
    }
  }

  /**
   * Peel a possibly unpeeled reference to an annotated tag.
   * <p>
   * If the ref cannot be peeled (as it does not refer to an annotated tag)
   * the peeled id stays null, but {@link Ref#isPeeled()} will be true.
   *
   * @param ref
   *            The ref to peel
   * @return <code>ref</code> if <code>ref.isPeeled()</code> is true; else a
   *         new Ref object representing the same data as Ref, but isPeeled()
   *         will be true and getPeeledObjectId will contain the peeled object
   *         (or null).
   */
  public Ref peel(final Ref ref) {
    try {
      return getRefDatabase().peel(ref);
    } catch (IOException e) {
      // Historical accident; if the reference cannot be peeled due
      // to some sort of repository access problem we claim that the
      // same as if the reference was not an annotated tag.
      return ref;
    }
  }

  /**
   * @return a map with all objects referenced by a peeled ref.
   */
  public Map<AnyObjectId, Set<Ref>> getAllRefsByPeeledObjectId() {
    Map<String, Ref> allRefs = getAllRefs();
    Map<AnyObjectId, Set<Ref>> ret = new HashMap<AnyObjectId, Set<Ref>>(allRefs.size());
    for (Ref ref : allRefs.values()) {
      ref = peel(ref);
      AnyObjectId target = ref.getPeeledObjectId();
      if (target == null)
        target = ref.getObjectId();
      // We assume most Sets here are singletons
      Set<Ref> oset = ret.put(target, Collections.singleton(ref));
      if (oset != null) {
        // that was not the case (rare)
        if (oset.size() == 1) {
          // Was a read-only singleton, we must copy to a new Set
          oset = new HashSet<Ref>(oset);
        }
        ret.put(target, oset);
        oset.add(ref);
      }
    }
    return ret;
  }

  /**
   * @return the index file location
   * @throws NoWorkTreeException
   *             if this is bare, which implies it has no working directory.
   *             See {@link #isBare()}.
   */
  public File getIndexFile() throws NoWorkTreeException {
    if (isBare())
      throw new NoWorkTreeException();
    return indexFile;
  }

  /**
   * Create a new in-core index representation and read an index from disk.
   * <p>
   * The new index will be read before it is returned to the caller. Read
   * failures are reported as exceptions and therefore prevent the method from
   * returning a partially populated index.
   *
   * @return a cache representing the contents of the specified index file (if
   *         it exists) or an empty cache if the file does not exist.
   * @throws NoWorkTreeException
   *             if this is bare, which implies it has no working directory.
   *             See {@link #isBare()}.
   * @throws IOException
   *             the index file is present but could not be read.
   * @throws CorruptObjectException
   *             the index file is using a format or extension that this
   *             library does not support.
   */
  public DirCache readDirCache() throws NoWorkTreeException,
      CorruptObjectException, IOException {
    return DirCache.read(getIndexFile(), getFS());
  }

  /**
   * Create a new in-core index representation, lock it, and read from disk.
   * <p>
   * The new index will be locked and then read before it is returned to the
   * caller. Read failures are reported as exceptions and therefore prevent
   * the method from returning a partially populated index.
   *
   * @return a cache representing the contents of the specified index file (if
   *         it exists) or an empty cache if the file does not exist.
   * @throws NoWorkTreeException
   *             if this is bare, which implies it has no working directory.
   *             See {@link #isBare()}.
   * @throws IOException
   *             the index file is present but could not be read, or the lock
   *             could not be obtained.
   * @throws CorruptObjectException
   *             the index file is using a format or extension that this
   *             library does not support.
   */
  public DirCache lockDirCache() throws NoWorkTreeException,
      CorruptObjectException, IOException {
    // we want DirCache to inform us so that we can inform registered
    // listeners about index changes
    IndexChangedListener l = new IndexChangedListener() {

      public void onIndexChanged(IndexChangedEvent event) {
        notifyIndexChanged();
      }
    };
    return DirCache.lock(getIndexFile(), getFS(), l);
  }

  static byte[] gitInternalSlash(byte[] bytes) {
    if (File.separatorChar == '/')
      return bytes;
    for (int i=0; i<bytes.length; ++i)
      if (bytes[i] == File.separatorChar)
        bytes[i] = '/';
    return bytes;
  }

  /**
   * @return an important state
   */
  public RepositoryState getRepositoryState() {
    if (isBare() || getDirectory() == null)
      return RepositoryState.BARE;

    // Pre Git-1.6 logic
    if (new File(getWorkTree(), ".dotest").exists())
      return RepositoryState.REBASING;
    if (new File(getDirectory(), ".dotest-merge").exists())
      return RepositoryState.REBASING_INTERACTIVE;

    // From 1.6 onwards
    if (new File(getDirectory(),"rebase-apply/rebasing").exists())
      return RepositoryState.REBASING_REBASING;
    if (new File(getDirectory(),"rebase-apply/applying").exists())
      return RepositoryState.APPLY;
    if (new File(getDirectory(),"rebase-apply").exists())
      return RepositoryState.REBASING;

    if (new File(getDirectory(),"rebase-merge/interactive").exists())
      return RepositoryState.REBASING_INTERACTIVE;
    if (new File(getDirectory(),"rebase-merge").exists())
      return RepositoryState.REBASING_MERGE;

    // Both versions
    if (new File(getDirectory(), Constants.MERGE_HEAD).exists()) {
      // we are merging - now check whether we have unmerged paths
      try {
        if (!readDirCache().hasUnmergedPaths()) {
          // no unmerged paths -> return the MERGING_RESOLVED state
          return RepositoryState.MERGING_RESOLVED;
        }
      } catch (IOException e) {
        // Can't decide whether unmerged paths exists. Return
        // MERGING state to be on the safe side (in state MERGING
        // you are not allow to do anything)
      }
      return RepositoryState.MERGING;
    }

    if (new File(getDirectory(), "BISECT_LOG").exists())
      return RepositoryState.BISECTING;

    if (new File(getDirectory(), Constants.CHERRY_PICK_HEAD).exists()) {
      try {
        if (!readDirCache().hasUnmergedPaths()) {
          // no unmerged paths
          return RepositoryState.CHERRY_PICKING_RESOLVED;
        }
      } catch (IOException e) {
        // fall through to CHERRY_PICKING
      }

      return RepositoryState.CHERRY_PICKING;
    }

    return RepositoryState.SAFE;
  }

  /**
   * Check validity of a ref name. It must not contain character that has
   * a special meaning in a Git object reference expression. Some other
   * dangerous characters are also excluded.
   *
   * For portability reasons '\' is excluded
   *
   * @param refName
   *
   * @return true if refName is a valid ref name
   */
  public static boolean isValidRefName(final String refName) {
    final int len = refName.length();
    if (len == 0)
      return false;
    if (refName.endsWith(".lock"))
      return false;

    int components = 1;
    char p = '\0';
    for (int i = 0; i < len; i++) {
      final char c = refName.charAt(i);
      if (c <= ' ')
        return false;
      switch (c) {
      case '.':
        switch (p) {
        case '\0': case '/': case '.':
          return false;
        }
        if (i == len -1)
          return false;
        break;
      case '/':
        if (i == 0 || i == len - 1)
          return false;
        components++;
        break;
      case '{':
        if (p == '@')
          return false;
        break;
      case '~': case '^': case ':':
      case '?': case '[': case '*':
      case '\\':
        return false;
      }
      p = c;
    }
    return components > 1;
  }

  /**
   * Strip work dir and return normalized repository path.
   *
   * @param workDir Work dir
   * @param file File whose path shall be stripped of its workdir
   * @return normalized repository relative path or the empty
   *         string if the file is not relative to the work directory.
   */
  public static String stripWorkDir(File workDir, File file) {
    final String filePath = file.getPath();
    final String workDirPath = workDir.getPath();

    if (filePath.length() <= workDirPath.length() ||
        filePath.charAt(workDirPath.length()) != File.separatorChar ||
        !filePath.startsWith(workDirPath)) {
      File absWd = workDir.isAbsolute() ? workDir : workDir.getAbsoluteFile();
      File absFile = file.isAbsolute() ? file : file.getAbsoluteFile();
      if (absWd == workDir && absFile == file)
        return "";
      return stripWorkDir(absWd, absFile);
    }

    String relName = filePath.substring(workDirPath.length() + 1);
    if (File.separatorChar != '/')
      relName = relName.replace(File.separatorChar, '/');
    return relName;
  }

  /**
   * @return true if this is bare, which implies it has no working directory.
   */
  public boolean isBare() {
    return workTree == null;
  }

  /**
   * @return the root directory of the working tree, where files are checked
   *         out for viewing and editing.
   * @throws NoWorkTreeException
   *             if this is bare, which implies it has no working directory.
   *             See {@link #isBare()}.
   */
  public File getWorkTree() throws NoWorkTreeException {
    if (isBare())
      throw new NoWorkTreeException();
    return workTree;
  }

  /**
   * Force a scan for changed refs.
   *
   * @throws IOException
   */
  public abstract void scanForRepoChanges() throws IOException;

  /**
   * Notify that the index changed
   */
  public abstract void notifyIndexChanged();

  /**
   * @param refName
   *
   * @return a more user friendly ref name
   */
  public static String shortenRefName(String refName) {
    if (refName.startsWith(Constants.R_HEADS))
      return refName.substring(Constants.R_HEADS.length());
    if (refName.startsWith(Constants.R_TAGS))
      return refName.substring(Constants.R_TAGS.length());
    if (refName.startsWith(Constants.R_REMOTES))
      return refName.substring(Constants.R_REMOTES.length());
    return refName;
  }

  /**
   * @param refName
   * @return a {@link ReflogReader} for the supplied refname, or null if the
   *         named ref does not exist.
   * @throws IOException the ref could not be accessed.
   */
  public abstract ReflogReader getReflogReader(String refName)
      throws IOException;

  /**
   * Return the information stored in the file $GIT_DIR/MERGE_MSG. In this
   * file operations triggering a merge will store a template for the commit
   * message of the merge commit.
   *
   * @return a String containing the content of the MERGE_MSG file or
   *         {@code null} if this file doesn't exist
   * @throws IOException
   * @throws NoWorkTreeException
   *             if this is bare, which implies it has no working directory.
   *             See {@link #isBare()}.
   */
  public String readMergeCommitMsg() throws IOException, NoWorkTreeException {
    if (isBare() || getDirectory() == null)
      throw new NoWorkTreeException();

    File mergeMsgFile = new File(getDirectory(), Constants.MERGE_MSG);
    try {
      return RawParseUtils.decode(IO.readFully(mergeMsgFile));
    } catch (FileNotFoundException e) {
      // MERGE_MSG file has disappeared in the meantime
      // ignore it
      return null;
    }
  }

  /**
   * Write new content to the file $GIT_DIR/MERGE_MSG. In this file operations
   * triggering a merge will store a template for the commit message of the
   * merge commit. If <code>null</code> is specified as message the file will
   * be deleted
   *
   * @param msg
   *            the message which should be written or <code>null</code> to
   *            delete the file
   *
   * @throws IOException
   */
  public void writeMergeCommitMsg(String msg) throws IOException {
    File mergeMsgFile = new File(gitDir, Constants.MERGE_MSG);
    if (msg != null) {
      FileOutputStream fos = new FileOutputStream(mergeMsgFile);
      try {
        fos.write(msg.getBytes(Constants.CHARACTER_ENCODING));
      } finally {
        fos.close();
      }
    } else {
      FileUtils.delete(mergeMsgFile, FileUtils.SKIP_MISSING);
    }
  }

  /**
   * Return the information stored in the file $GIT_DIR/MERGE_HEAD. In this
   * file operations triggering a merge will store the IDs of all heads which
   * should be merged together with HEAD.
   *
   * @return a list of commits which IDs are listed in the MERGE_HEAD
   *         file or {@code null} if this file doesn't exist. Also if the file
   *         exists but is empty {@code null} will be returned
   * @throws IOException
   * @throws NoWorkTreeException
   *             if this is bare, which implies it has no working directory.
   *             See {@link #isBare()}.
   */
  public List<ObjectId> readMergeHeads() throws IOException, NoWorkTreeException {
    if (isBare() || getDirectory() == null)
      throw new NoWorkTreeException();

    byte[] raw = readGitDirectoryFile(Constants.MERGE_HEAD);
    if (raw == null)
      return null;

    LinkedList<ObjectId> heads = new LinkedList<ObjectId>();
    for (int p = 0; p < raw.length;) {
      heads.add(ObjectId.fromString(raw, p));
      p = RawParseUtils
          .nextLF(raw, p + Constants.OBJECT_ID_STRING_LENGTH);
    }
    return heads;
  }

  /**
   * Write new merge-heads into $GIT_DIR/MERGE_HEAD. In this file operations
   * triggering a merge will store the IDs of all heads which should be merged
   * together with HEAD. If <code>null</code> is specified as list of commits
   * the file will be deleted
   *
   * @param heads
   *            a list of commits which IDs should be written to
   *            $GIT_DIR/MERGE_HEAD or <code>null</code> to delete the file
   * @throws IOException
   */
  public void writeMergeHeads(List<ObjectId> heads) throws IOException {
    writeHeadsFile(heads, Constants.MERGE_HEAD);
  }

  /**
   * Return the information stored in the file $GIT_DIR/CHERRY_PICK_HEAD.
   *
   * @return object id from CHERRY_PICK_HEAD file or {@code null} if this file
   *         doesn't exist. Also if the file exists but is empty {@code null}
   *         will be returned
   * @throws IOException
   * @throws NoWorkTreeException
   *             if this is bare, which implies it has no working directory.
   *             See {@link #isBare()}.
   */
  public ObjectId readCherryPickHead() throws IOException,
      NoWorkTreeException {
    if (isBare() || getDirectory() == null)
      throw new NoWorkTreeException();

    byte[] raw = readGitDirectoryFile(Constants.CHERRY_PICK_HEAD);
    if (raw == null)
      return null;

    return ObjectId.fromString(raw, 0);
  }

  /**
   * Write cherry pick commit into $GIT_DIR/CHERRY_PICK_HEAD. This is used in
   * case of conflicts to store the cherry which was tried to be picked.
   *
   * @param head
   *            an object id of the cherry commit or <code>null</code> to
   *            delete the file
   * @throws IOException
   */
  public void writeCherryPickHead(ObjectId head) throws IOException {
    List<ObjectId> heads = (head != null) ? Collections.singletonList(head)
        : null;
    writeHeadsFile(heads, Constants.CHERRY_PICK_HEAD);
  }

  /**
   * Read a file from the git directory.
   *
   * @param filename
   * @return the raw contents or null if the file doesn't exist or is empty
   * @throws IOException
   */
  private byte[] readGitDirectoryFile(String filename) throws IOException {
    File file = new File(getDirectory(), filename);
    try {
      byte[] raw = IO.readFully(file);
      return raw.length > 0 ? raw : null;
    } catch (FileNotFoundException notFound) {
      return null;
    }
  }

  /**
   * Write the given heads to a file in the git directory.
   *
   * @param heads
   *            a list of object ids to write or null if the file should be
   *            deleted.
   * @param filename
   * @throws FileNotFoundException
   * @throws IOException
   */
  private void writeHeadsFile(List<ObjectId> heads, String filename)
      throws FileNotFoundException, IOException {
    File headsFile = new File(getDirectory(), filename);
    if (heads != null) {
      BufferedOutputStream bos = new SafeBufferedOutputStream(
          new FileOutputStream(headsFile));
      try {
        for (ObjectId id : heads) {
          id.copyTo(bos);
          bos.write('\n');
        }
      } finally {
        bos.close();
      }
    } else {
      FileUtils.delete(headsFile, FileUtils.SKIP_MISSING);
    }
  }
}
TOP

Related Classes of org.eclipse.jgit.lib.Repository

TOP
Copyright © 2018 www.massapi.com. All rights reserved.
All source code are property of their respective owners. Java is a trademark of Sun Microsystems, Inc and owned by ORACLE Inc. Contact coftware#gmail.com.