Need to change old commit (and regenerate tree)

Hi all

My question is this: I have two branches (say B and C) where one is reachable from the other (say B is ancestor of C), and if they are separately rebased about/onto the same point, why do B and C become non-overlapping branches?

Let me explain with an example.

Say I have the following commit line:

A--------o--------o--------o--------B--------o--------o--------C

A, B and C are branches (so that B is reachable from C, and A is reachable from B). [For ease, I am drawing the branches at the same level since there are no real diverging branches here.]

I then realise that I want to change the commit A and have both B and C rebased on this changed commit.

Now when I do a
  $ git rebase --onto A2 A C

This results in two parallel trees like these:

A--------o--------o--------o--------B--------o--------o--------o(old C)

A2--------o--------o--------o--------o--------o--------o--------C

Now I go about rebasing B. I can of course 'reset' B to C~3. But
alternatively, if I decide to do a rebase:
  $ git rebase --onto A2 A B

I will end up getting

A--------o--------o--------o--------o(old B)--------o--------o--------o(old C)

A2--------o--------o--------o--------o--------o--------o--------C
   \
     ` -----o--------o--------o--------B

Instead of (what I initially expected):

A--------o--------o--------o--------o(old B)--------o--------o--------o(old C)

A2--------o--------o--------o--------B--------o--------o--------C

So what I am missing here? Aren't the new commits B~1, B~2, B~3 identical to C~4, C~5, C~6 (respectively) in all ways so as to have gotten them the same SHA1 and hence appear as what I expected them to appear?

I have taken a simple example here. In reality, I wanted to change a not so new commit (on the main line), and there were many branches diverging out from the main line after the (bad) commit. I initially thought I could just write a simple script that would rebase all branches that have the bad commit as an ancestor, and the new tree would be a mirror image of the original. But that was not to be! There is also the problem of resetting tags (and possible notes).

I am sorry if this has already been discussed, please point me to the right resource if so.

My git version:
  $ git --version
  git version 1.7.0.5

Thanks Antriksh Pany

Antriksh Pany <antriksh.pany@gmail.com> writes:

No, they have a different commit time, which is also part of the hash.

Andreas.

On Fri, May 21, 2010 at 8:09 AM, Andreas Schwab <schwab@linux-m68k.org> wrote:

Of course, even if the commit time was forged to be the same, the parent of B~3 is different to the parent of C~6 and since the parent is also contributes bits to the respective hashes, B~3 will necessarily (unlikely hash collisions excepted!) have a different hash to C~6

jon.

Hi Jon

I guess the parent of both B~3 as well as C~6 is A2. So if, as you say, the time can be made identical, they should yield the same SHA1 IMHO.

On Fri, May 21, 2010 at 12:05 AM, Jon Seymour <jon.seymour@gmail.com> wrote:

On Fri, May 21, 2010 at 00:09, Andreas Schwab <schwab@linux-m68k.org> wrote:

Indeed.

To avoid this, you have to:
  - rebase B on top of A2 first,

        git rebase --onto A2 A B

  - rebase of C on top of the new B.

        git rebase --onto B B_old C
       ("git rebase --onto B A C" should work too, as usually git is
smart enough to see
        that A-B_old is already applied. Use "git rebase --skip" if it isn't)

If A is an ancestor of A2, you can simplify to:

    git rebase A2 B
    git rebase B C

(Disclaimer: the examples without --onto I use almost daily, the ones with I don't)

Gr{oetje,eeting}s,

						Geert

-- Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
							    -- Linus Torvalds

Thanks a lot.

I guess it is then not so straightforward to regenerate trees with their relative structure intact when an old commit changes. I was initially of the opinion that it would be a trivial rebasing of all branches from which the commit was reachable. Apparently, not so.

On Fri, May 21, 2010 at 6:31 AM, Geert Uytterhoeven <geert@linux-m68k.org> wrote: