Sanitizing a branch (I got a result)

[Erik de Castro Lopo]

Erik de Castro Lopo wrote:

> When I come up with something I'm happy with  I'll report back.

Well here I am. 

After reading way more documentation than I really like to and a few
false starts I followed the route sugggested by Stephen J. Turnbull:

> So export a fastimport database, import to git, do the history etc
> munging in git using filter-branch, then import back to bzr.

I expected to have to do this a couple of times to get it right, but
at the end of my first trial run I was happy enough with the result
that I decided to keep it.

I was using the standard bzr that comes with Ubuntu Intrepid (1.6.1)
with the fast-import plugin from launchpad. I did the fast-export part
to a file, mainly so I could have a look at what the file contained :-).
This process was completely painless.

Importing into  Git was also painless (and fast) but resulted in a
directory that was empty except for a .git directory. A visit to the
Git IRC channel and I ran "git reset --hard HEAD" to populate the
directory.

I then 6 passes removing cruft files using:

  git filter-branch --tree-filter 'rm -f file1 file2' HEAD

The "-f" in the rm is important so that it doesn't fail in earlier
revisions where the files may not exist. Each of these passes through
my tree took about 10 minutes.

Once I was happy with the final result "bzr fast-import" gave me my
new sanitized tree (after making me wait for over an hour :-) ).

I now intend to use bzr-loom to add my cruft back into the tree in
a more controlled manner. I will then manually rebase a couple of
related hacking branches on top of the new branch.

I must say I am very impressed with how well Bzr and Git interacted
here. I was also impressed with the speed of Git.

Cheers,
Erik