Re: Issue with global config defaults "user.useConfigOnly = true" + "pull.rebase = preserve" - "user.email"

From: Junio C Hamano <gitster@pobox.com>
Date: Jul 29, 2016, 17:47 UTC
Message-ID: <xmqqoa5gnxow.fsf@gitster.mtv.corp.google.com>
In-Reply-To: <CAG0BQXnVAYdpk9EM_uiD+=UKSKmK=z1YEar5MresTr5XfDCxHw@mail.gmail.com>

Dakota Hawkins <dakotahawkins@gmail.com> writes:

Show 14 quoted lines

> I have a question which may be a bug (I'm a bit skeptical), but here goes:
>
> In my global .gitconfig, I have "user.useConfigOnly = true" and
> user.email isn't set there (I prefer to be forced to set it on a
> per-repo basis, as I use different emails for work and personal
> repos). I ALSO have "pull.rebase = preserve" set.
>
> An example of the problem I have is with tools like golang (I filed an
> issue there, they closed it and suggested the problem is with git or
> my config: https://github.com/golang/go/issues/16516#issuecomment-235800085)
> that use git to pull in package repos without any real user
> interaction. When something like that runs a git pull for me (to
> update a package repo) my global config makes it try to rebase, which
> fails because git doesn't know who I am.

It's an interesting chicken-and-egg problem that user.useConfigOnly introduces. It seems that the design of that configuration variable is not perfect and has room for improvement.

Show 9 quoted lines

> In those cases specifically, I never have local commits that differ
> from the remote, so a "pull --ff-only" should leave me in the same
> state as a "pull --rebase".
>
> Is this a case of rebase trying to make sure it has enough information
> for me to be a committer before knowing whether I even need to rewrite
> any commits, and could/should that be avoided?  Alternatively (or also)
> could/should rebase detect that a fast-forward is possible and prefer
> to do that instead?

I think that is a reasonable argument, but to solve this for a more general case, shouldn't we be discussing a solution that would also work when rebase _does_ need to create a new commit? And when the latter is solved, I would imagine that "this rebase happens to be fast-forward, and not having an ident shouldn't be an issue for this special case" would become moot.