« »

Saturday, 10 July 2010

My quick dirty hack to move a (small) part of a Mercurial (hg) repo to Git. In my case this was for moving my single file “testlib.py” from here on bitbucket to here on github

1. Dump the log of that part of the hg repo to a file.

    export WORKDIR=$HOME/tmp/migrate
    cd HGREPO/foo
    hg log -pv . > $WORKDIR/full.patch
   

2. Create the starter git repo

    cd $WORKDIR
    git init foo
   

3. Break up the log into a number of “changesetNNN.patch” files with this Python code:

    # parse.py
    import codecs
    changeset = []
    i = 0
       
    def write_changeset():
        global changeset
        if not changeset:
            return
        codecs.open("changeset%05d.patch" % i, 'w', "utf-8").write(''.join(changeset))
        i += 1
        changeset = []
       
    for line in open("full.patch"):
        if line.startswith("changeset:"):
            write_changeset()
        changeset.append(line)
    write_changeset()
   

   and then run:

    python parse.py
   

4. Apply and commit each patch with this Python script

    # Usage: python apply.py TARGET-REPO-BASE-DIR
    import os
    from os.path import *
    from glob import glob
    import subprocess
    from pprint import pprint
       
    def apply_patch(target_repo_base_dir, patch_path):
        content = open(patch_path).read()
        header, diff = content.split('\n\n\n', 1)
        assert diff.startswith("diff --git a")
        fields = {}
        lines = header.splitlines(False)
        for i, line in enumerate(lines):
            key, value = line.split(':', 1)
            if key == "description":
                fields[key] = '\n'.join(lines[i+1:])
                break
            value = value.strip()
            fields[key] = value
        # Do any path renamings here. For example, I wanted to move from
        # "testlib/testlib.py" in the old repo to "lib/testlib.py" in the new.
        diff = diff.replace('a/testlib/testlib.py', 'a/lib/testlib.py')
        diff = diff.replace('b/testlib/testlib.py', 'b/lib/testlib.py')
        f = open(patch_path+".diff", 'w')
        f.write(diff)
        f.close()
        f = open(patch_path+".msg", 'w')
        f.write(fields["description"])
        f.close()
        cwd = target_repo_base_dir
        subprocess.check_call(['git', 'apply', '--whitespace=nowarn',
            abspath(patch_path+".diff")], cwd=cwd)
        subprocess.check_call(['git', 'add', 'lib/testlib.py'], cwd=cwd)
        subprocess.check_call(['git', 'commit', '--date', fields["date"],
            '-F', abspath(patch_path+".msg")], cwd=cwd)
       
    if __name__ == "__main__":
        target_repo_base_dir = sys.argv[1]
        patches = list(sorted(glob("changeset*.patch")))
        for patch_path in patches:
            print "--", patch_path
            apply_patch(target_repo_base_dir, patch_path)
   

   Then run:

    python apply.py foo    # apply all changeset*.patch files to "foo" git repo
   

Now you can push this Git repo to github or whereever.

To improve on

• You currently need to manually create the dir structure first.
• This doesn’t currently used the parse “user” field from the hg commit log for the “git commit -a AUTHOR” command. Mainly this is because I didn’t need that, but also because the “user” value in the hg log isn’t the configured full user name and email, but just the short username. Maybe that was just me.