From: David <bouncingcats@gmail.com>
Date: Sat, 03 Jul 2010 13:38:14 GMT
Subject: *Non*-interactive squash a range ?
Message-ID: <AANLkTimTx2dVlK9KKi1nZpOadwDcfyzhE8nffCyK1Sqm@mail.gmail.com>

Thanks for the awesome git. I have a basic question. I am subscribed
to this list.

Having successfully used rebase -i to squash, I wonder about some
method to *non*-interactively squash a sequence of intermediate
commits. On a local experimental branch.

Lets say the starting point is this branch named "uglybranch" with HEAD=J

    E---F---G---H---I---J  <uglybranch>

where commit G added a previously untracked file.

My desired result is this:

      ------------------K  <cleanbranch>
     /
    E---F---G---H---I---J  <uglybranch>

where K and J are identical worktrees.

I've used rebase -i as follows:
 $ git checkout uglybranch
 $ git checkout -b cleanbranch
 $ git rebase -i E
 # used interactive editor to squash all F..J onto E to produce K, on
cleanbranch only, uglybranch is not changed.

This gives exactly the result I want, in particular the file added at
G is preserved, but the drawback is that the method is needlessly
interactive. I want a foolproof non-interactive method that achieves
the same result. Namely squashing *all* the commits in the range F..J

I tried this which seems to work:
 $ git checkout uglybranch
 $ git checkout -b cleanbranch
 $ git reset --mixed E          # worktree same as uglybranch
 $ git add .                    # adds any untracked files
 $ git commit -a                # -a detects any removed files

So I'm thinking of using that method in a script. Have I overlooked
anything? Is there a better way?