Re: [GIT PULL] Ceph updates for 4.6-rc1

[Linus Torvalds]

On Sat, Mar 26, 2016 at 2:46 PM, Ilya Dryomov <idryomov@xxxxxxxxx> wrote:
>
> FWIW, it matches -next content-wise.  SHAs don't match because of an
> extra Signed-off-by in an unrelated commit, outside of fs/ceph.

That's how git works. After you rebase, SHA's will *never* match,
because you changed the history (dates, parenthood etc).

The whole "it matches content-wise" is immaterial. You rebased and now
I can see that it's not what was in next. It's really that simple.

Just as an example of what rebasing does: it means that when I look at
"ok, what is in next but not in my tree yet" to see what might be
pending, the rebased commits don't cancel out, because they are
different commits.

Yes, yes, I've been lax about this. I have been ranting against people
rebasing for years, but in the end I've usually let it slide.

But dammit, when things come in at the very end of the merge window,
which makes me grumpy to begin with, AND it is then re-based, at that
point I'm not inclined to let things slide any more.

At that point you guys are actively working to piss me off, and that
means that I'm not in the least in the mood of pulling your work.

Why *was* the pull request sent at the very end of the merge window
anyway?  Was the code not ready?

So let me say again: keep your *own* tree in good shape.

That's actually the only really valid reason for rebasing: if your
*own* tree has something horribly bad going on, like majorly messed up
commit messages, or history that is completely broken and will cause
problems for people who want to bisect bugs, or things like that. Then
"git rebase" is a perfectly good thing to fix bad things that are in
_your_ tree.

But no, git rebase is not a "let's react to random things that
happened in other peoples trees". At that point you're worrying about
entirely the wrong thing.

It's doubly wrong when you rebase on top of some random point in the
middle of the merge window. That "middle of the merge window" is
pretty much by definition not a good stable base for development.
Maybe this merge window has been good and we haven't had any real
problems, but how can you be sure?

So even _if_ you have to rebase because of some major mess-up in the
tree, rebasing on top of "random commit in the merge window" is one of
the last things you should do.

                   Linus