CVS to git: parsing ChangeLog entries?

Hello,

when migrating a project that uses GNU-style ChangeLogs from CVS to git, is it possible to extract patch author information from the ChangeLog entries rather than from the CVS commit logs? For simplicity let's first assume the project used only one ChangeLog file.

Asking because it is not uncommon that patches are committed on behalf of other people, and it would be nice to credit them.

Related question: when CVS commit logs have varying encoding, say, some latin1 and some UTF-8, is it possible to have uniformly encoded git log entries?

Thank you, and please Cc: me on replies, Ralf

On Wed, Mar 12, 2008 at 08:08:27PM +0100, Ralf Wildenhues wrote:

I don't think there is a way to do this automatically with git-cvsimport. However, once imported, I think you could rewrite history using git-filter-branch with a filter that looked at the diff of ChangeLog for that commit and rewrote the author. See the documentation for git-filter-branch.

I don't think git-cvsimport does much with encodings at all. But again, you could probably go back through the imported repo with git-filter-branch and iconv the commit messages as appropriate.

-Peff

* Jeff King wrote on Thu, Mar 13, 2008 at 06:03:22PM CET:

Thank you, I just learned a new cool tool!

FWIW, here's what I used successfully on a repository with only one ChangeLog file (some fiddling to cope with format variations):

git filter-branch -d /dev/shm/t --tree-filter '
  line=`sed -n "s,^[12][90][0-9][0-9]-[0-1][0-9]-[0-3][0-9]  *\([A-Za-z].*\),\1,;
                s,.*  \([A-Za-z].*\),\1,;
                /./{
                        s/(tiny change)//
                        s/ *$//
                        p
                        q
                }" ChangeLog`
  author=`echo "$line" | sed "s, *[<(].*,,"`
  email=`echo "$line" | sed "s,[^,(]*[<(],<,; s/[)>].*/>/"`
  test -n "$author" && test -n "$email" && {
        GIT_AUTHOR_NAME="$author"
        GIT_AUTHOR_EMAIL="$email"
        export GIT_AUTHOR_NAME GIT_AUTHOR_EMAIL
  }'

I'll try that, too.

Thanks! Ralf

I'll answer your questions from the point of view of cvs2git, a.k.a. cvs2svn (http://cvs2svn.tigris.org).

Ralf Wildenhues wrote:

There is no builtin support for this in cvs2git. But the place where the author is determined knows about the whole changeset, including any simultaneous changes to the changelog. So it should be possible to add this functionality without too much work.

cvs2git allows you to specify multiple encodings. It tries one after the other until one works successfully. It also has hooks where you can add your own decoder using arbitrary Python code.

That reminds me that there is a Python universal decoder that uses heuristics to determine the encoding of an arbitrary octet stream. That might be a nice thing to add support for....

Michael