It could be rebased upstream. While it's a really bad practice, it could happen.

Ah ok, I think I understand the concern. So 'git fetch' and 'git merge' explicitly will stop an upstream rebase from making destructive changes to the history of your local branch(?)

As you say it's bad practice and shouldn't happen. I use gitlab and GitHub where branches can be protected to only allow merges via their web interfaces.

GitLab CEO here, in GitLab you can protect branches to not be rebased at all (web or command line), see https://about.gitlab.com/2014/11/26/keeping-your-code-protec...

I'm not aware that GitHub offers the same functionality.

On the page where the ssh finger print is listed, can you list the RSA finger print too?

I am not well versed in working with git over ssh, but after following the instructions on GitLab to generate a key, then git would keep showing me the RSA finger print which doesn't match the one provided on the website. After looking around a bit, it looks like the one listed is ECDSA.

tl;dr: It was late. Finger prints didn't match. Only logical assumption was NSA. :P (Not really, obviously, but the finger prints could be more clear.)

Thanks, fixed in https://gitlab.com/gitlab-com/www-gitlab-com/commit/2493f208...

Also, we added all other keys: https://gitlab.com/gitlab-com/www-gitlab-com/commit/79bdb63f...

but you still damage your local copy and it's a pain to clean up, specially if you had local changes/commits

The idea is not to have upstream rebases so you can't pull them by accident.

I don't think git pull would ever destroy your history (AFAIK it's exactly the same as a fetch+merge), at worst you end up in a merge conflict, but it's easily revertable. However after a git fetch and a git log you can see beforehand if there is anything nasty going on.

Nothing unless you're rebasing upstream, but I think the point he was trying to make was that you should always run `git fetch` and then `merge` from there (again, there are some situations where the pull is fine IMHO).

git pull just runs git fetch then git merge.

At least that is what the docs say: http://git-scm.com/docs/git-pull

What am I missing?

but it tries to be smart about how to merge. and usually it's not what you want. i consider it the clippy of linus.

if you run merge/rebase/etc on its own, it will tell and ask you about anything out of the ordinary and then there's no surprise.

again, it's all about avoiding errors on the rare cases because you developed dangerous finger memory.

Beat me to it.

Why?

nothing. but tomorrow, when you are used to type git pull, you will run it on a branch with changes and you have no idea if it will rebase, merge, fast forward... and /then/ you will see the problem. (and open yet another stackoverflow question on how to undo the last merge :)