Re: Looks like --amend create commit... don't understand

[Matthieu Moy <Matthieu.Moy@grenoble-inp.fr>]

mat <matthieu.stigler@gmail.com> writes:

> commits. But when looking on gitk, I see that
> git commit blabla --amend
>
> does create each and every time a new commit,

Yes. In Git, objects (including commits) are fundamentally immutable.
You never modify an object, you create a new one, and usually forget
about the old one.

> I mean: with git log,
> there is only one commit, but on gitk, I see many, with the same name
> but different revision ID....

The revision ID is the sha1sum of the commit content. So, if they have
different contents, they _must_ have different IDs.

What happens is that you start with a history like

A---B <- master

and then try to --amend B. What happens is that you create a new B,
say B', and let the branch you sit on point to this new one. Like:

A---B
 \
  `-B' <- master

Most of the time, B has just become unreachable: no branch contain it,
no tag point to it, ... so you actually see

A---B' <- master

and it very much looks like you just modified B.

Now, if you made B reachable before your commit --amend, like by
pushing it to another repo, letting someone else pull from it, or if B
is already part of another branch, then the situation is different.

For example, if B is already part of a branch, like this:

    .---- temp-branch
    v
A---B---C <- master

If you do a checkout of temp-branch, then commit --amend, you end up
in this situation :

A---B---C <- master
 \
  `-B' <- temp-branch

The solutions:

* If you already published your commit, then don't amend it. Period.

* If you want to rewrite an old commit in a branch, then read about
  "git rebase -i", but make sure you read all the warnings about
  rebase before you do so.

-- 
Matthieu Moy
http://www-verimag.imag.fr/~moy/