views:

2353

answers:

3

I want to modify a commit message deeper in history. And I've pushed many new commits. How do I chnage the commit message? Is it possible?

Thanks!

+12  A: 

The message from Linus Torvalds may answer your question:

Modify/edit old commit messages

Short answer: you can not (if pushed).


extract (Linus refers to BitKeeper as BK):

Side note, just out of historical interest: in BK you could.

And if you're used to it (like I was) it was really quite practical. I would apply a patch-bomb from Andrew, notice something was wrong, and just edit it before pushing it out.

I could have done the same with git. It would have been easy enough to make just the commit message not be part of the name, and still guarantee that the history was untouched, and allow the "fix up comments later" thing.

But I didn't.

Part of it is purely "internal consistency". Git is simply a cleaner system thanks to everything being SHA1-protected, and all objects being treated the same, regardless of object type. Yeah, there are four different kinds of objects, and they are all really different, and they can't be used in the same way, but at the same time, even if their encoding might be different on disk, conceptually they all work exactly the same.

But internal consistency isn't really an excuse for being inflexible, and clearly it would be very flexible if we could just fix up mistakes after they happen. So that's not a really strong argument.

The real reason git doesn't allow you to change the commit message ends up being very simple: that way, you can trust the messages. If you allowed people to change them afterwards, the messages are inherently not very trustworthy.


To be complete, you could rewrite your local commit history in order to reflect what you want, as suggested by sykora (with some rebase and reset --hard, gasp!)

However, once you publish your revised history again (with a git push origin +master:master, the + sign forcing the push to occur, even if it doesn't result in a "fast-forward" commit)... you might get into some trouble.

Extract from this other SO question:

I actually once pushed with --force to git.git repository and got scolded by Linus BIG TIME. It will create a lot of problems for other people. A simple answer is "don't do it".

VonC
+1  A: 

(From http://git.or.cz/gitwiki/GitTips#head-9f87cd21bcdf081a61c29985604ff4be35a5e6c0)

How to change commits deeper in history

Since history in Git is immutable, fixing anything but the most recent commit (commit which is not branch head) requires that the history is rewritten from the changed commit and forward.

You can use StGIT for that, initialize branch if necessary, uncommitting up to the commit you want to change, pop to it if necessary, make a change then refresh patch (with -e option if you want to correct commit message), then push everything and stg commit.

Or you can use rebase to do that. Create new temporary branch, rewind it to the commit you want to change using git reset --hard, change that commit (it would be top of current head), then rebase branch on top of changed commit, using git rebase --onto .

Or you can use git rebase --interactive, which allows various modifications like patch re-ordering, collapsing, ...

I think that should answer your question. However, note that if you have pushed code to a remote repository and people have pulled from it, then this is going to mess up their code histories, as well as the work they've done. So do it carefully.

sykora
Good answer in theory, mighty dangerous in practice: see http://stackoverflow.com/questions/253055#432518
VonC
A: 

This article is about your question.

Peter
Yes, but this article does specify: "as long as you haven’t pushed upstream yet", and I believe here the commit to modified has been pushed.
VonC
Furthermore, this article involves commit --amend, and being above 10K rep, I can see two other answers (with commit --amend) have been posted before being quickly deleted ;)
VonC
I didn't say, that this article solves the problems, but there this problem has been discussed already.
Peter
"this problem" ? I would merely point out the article mentionned by your link does not address the problem presented in the question at the top of this page.
VonC