How to show changes between commits with JGit

I am trying to show a git diff between two commits for a file. Basically, I did it as in https://github.com/centic9/jgit-cookbook/blob/master/src/main/java/org/dstadler/jgit/porcelain/ShowChangedFilesBetweenCommits.java

You can see my full code under https://github.com/svenhornberg/JGitDiff

public class RepoDiff {

    public void compare(byte[] fileContentOld, byte[] fileContentNew) {
        try {
            Repository repository = createNewRepository();
            Git git = new Git(repository);

            // create the file
            commitFileContent(fileContentOld, repository, git, true);
            commitFileContent(fileContentNew, repository, git, false);

            // The {tree} will return the underlying tree-id instead of the commit-id itself!
            ObjectId oldHead = repository.resolve("HEAD^^^^{tree}");   //here is my nullpointer
            ObjectId head = repository.resolve("HEAD^{tree}");

            System.out.println("Printing diff between tree: " + oldHead + " and " + head);

            // prepare the two iterators to compute the diff between
            ObjectReader reader = repository.newObjectReader();
            CanonicalTreeParser oldTreeIter = new CanonicalTreeParser();
            oldTreeIter.reset(reader, oldHead);
            CanonicalTreeParser newTreeIter = new CanonicalTreeParser();
            newTreeIter.reset(reader, head);

            // finally get the list of changed files
            List<DiffEntry> diffs= new Git(repository).diff()
                            .setNewTree(newTreeIter)
                            .setOldTree(oldTreeIter)
                            .call();
            for (DiffEntry entry : diffs) {
                System.out.println("Entry: " + entry);
            }
            System.out.println("Done");

        } catch (Exception exception) {
            exception.printStackTrace();
        }
    }

/**
 * Adds and optionally commits fileContent to a repository
 * @param commit True if file should be committed, False if not
 * @throws Exception catch all for testing
 */
private void commitFileContent(byte[] fileContent, Repository repository, Git git, boolean commit) throws Exception {

    File myfile = new File(repository.getDirectory().getParent(), "testfile");
    myfile.createNewFile();

    FileOutputStream fos = new FileOutputStream (myfile); 
    ByteArrayOutputStream baos = new ByteArrayOutputStream();
    baos.write(fileContent);
    baos.writeTo(fos);

    // run the add
    git.add().addFilepattern("testfile").call();

    if(commit) {
        // and then commit the changes
        git.commit().setMessage("Added fileContent").call();
    }

    fos.close();
}

/**
 * Helperfunction from
 * https://github.com/centic9/jgit-cookbook
 */
public static Repository createNewRepository() throws IOException {
    // prepare a new folder
    File localPath = File.createTempFile("TestGitRepository", "");
    localPath.delete();

    // create the directory
    Repository repository = FileRepositoryBuilder.create(new File(
            localPath, ".git"));
    repository.create();

    return repository;
    }
}

the code results in this message:

Printing diff between tree: null and AnyObjectId[c11a3a58c23b0dd6e541c4bcd553197772626bc6]
java.lang.NullPointerException
at  org.eclipse.jgit.internal.storage.file.UnpackedObjectCache$Table.index(UnpackedObjectCache.java:146)
at org.eclipse.jgit.internal.storage.file.UnpackedObjectCache$Table.contains(UnpackedObjectCache.java:109)
at org.eclipse.jgit.internal.storage.file.UnpackedObjectCache.isUnpacked(UnpackedObjectCache.java:64)
at org.eclipse.jgit.internal.storage.file.ObjectDirectory.openObject(ObjectDirectory.java:367)
at org.eclipse.jgit.internal.storage.file.WindowCursor.open(WindowCursor.java:145)
at org.eclipse.jgit.treewalk.CanonicalTreeParser.reset(CanonicalTreeParser.java:202)
at javadiff.RepoDiff.compare(RepoDiff.java:40)
at javadiff.GitDiff.main(GitDiff.java:30)

the following line must be false, but it is like in the example from jgit-cookbook

 ObjectId oldHead = repository.resolve("HEAD^^^^{tree}");   //here is my nullpointer

I tested HEAD^1{Tree} but this does not work, it looks like {tree} must be in the string for resolving a tree instead of a commit id. Any ideas to get it working?

Solution 1:

To obtain the tree of the head commit, call

git.getRepository().resolve( "HEAD^{tree}" )

and to obtain the tree of the parent of the HEAD commit, call

git.getRepository().resolve( "HEAD~1^{tree}" )

Search for 'Git caret and tilde' if you are interested in more details.

To summarize, here goes a snippet that computes the diff of two commits:

File file = new File( git.getRepository().getWorkTree(), "file.txt" );
writeFile( file, "first version" );
RevCommit newCommit = commitChanges();
writeFile( file, "second version" );
RevCommit oldCommit = commitChanges();

ObjectReader reader = git.getRepository().newObjectReader();
CanonicalTreeParser oldTreeIter = new CanonicalTreeParser();
ObjectId oldTree = git.getRepository().resolve( "HEAD^{tree}" ); // equals newCommit.getTree()
oldTreeIter.reset( reader, oldTree );
CanonicalTreeParser newTreeIter = new CanonicalTreeParser();
ObjectId newTree = git.getRepository().resolve( "HEAD~1^{tree}" ); // equals oldCommit.getTree()
newTreeIter.reset( reader, newTree );

DiffFormatter df = new DiffFormatter( new ByteArrayOutputStream() ); // use NullOutputStream.INSTANCE if you don't need the diff output
df.setRepository( git.getRepository() );
List<DiffEntry> entries = df.scan( oldTreeIter, newTreeIter );

for( DiffEntry entry : entries ) {
  System.out.println( entry );
}

private RevCommit commitChanges() throws GitAPIException {
  git.add().addFilepattern( "." ).call();
  return git.commit().setMessage( "commit message" ).call();
}

private static void writeFile( File file, String content ) throws IOException {
  FileOutputStream outputStream = new FileOutputStream( file );
  outputStream.write( content.getBytes( "UTF-8" ) );
  outputStream.close();
}

For further considerations about showing changes between commits, you may want to read this in-depth discussion of JGit's diff APIs that can be found here: http://www.codeaffine.com/2016/06/16/jgit-diff/

Solution 2:

I am using the following code to print diffs between two commits. Using DiffFormatter.scan seems simpler than the other solutions I have seen on SO.

It might be that I am missing some case(s) where this is not supported, but it works fine for our use case.

public static void main(String[] args) throws Exception {
    Repository repository = new FileRepositoryBuilder()
            .setGitDir(new File("c:/temp/jgit-test/.git")).build();
    // Here we get the head commit and it's first parent.
    // Adjust to your needs to locate the proper commits.
    RevCommit headCommit = getHeadCommit(repository);
    RevCommit diffWith = headCommit.getParent(0);
    FileOutputStream stdout = new FileOutputStream(FileDescriptor.out);
    try (DiffFormatter diffFormatter = new DiffFormatter(stdout)) {
        diffFormatter.setRepository(repository);
        for (DiffEntry entry : diffFormatter.scan(diffWith, headCommit)) {
            diffFormatter.format(diffFormatter.toFileHeader(entry));
        }
    }
}

private static RevCommit getHeadCommit(Repository repository) throws Exception {
    try (Git git = new Git(repository)) {
        Iterable<RevCommit> history = git.log().setMaxCount(1).call();
        return history.iterator().next();
    }
}