Re: [PATCH v2 02/14] pull: improve default warning

From: Felipe Contreras <felipe.contreras@gmail.com>
Date: Dec 7, 2020, 22:14 UTC
Message-ID: <CAMP44s3FRkdS9j=B0ht0Uviy4XE43PbObjXRk7hekRGPyoVCpg@mail.gmail.com>
In-Reply-To: <xmqqlfe99yvy.fsf@gitster.c.googlers.com>

On Mon, Dec 7, 2020 at 1:53 PM Junio C Hamano <gitster@pobox.com> wrote:

Show 10 quoted lines

>
> Felipe Contreras <felipe.contreras@gmail.com> writes:
>
> > They start by saying the same thing. But one errors out and says the
> > user must choose, and the other warns that in the future the user must
> > choose.
>
> Then I do not see the point in giving the warning---even in the
> future they do not have to choose as long as they are merely
> following along.

They don't. With my patch series they see the warning only when the pull is non-fast-forward.

Show 21 quoted lines

> >> > Just to put this series in context: it's only part 1; it does not
> >> > introduce pull.mode, and it doesn't make --ff-only the default.
> >>
> >> I'd view the "in a non-fast-forward situation, the warning kicks in
> >> to those who haven't chosen between merge and rebase (i.e. no
> >> pull.rebase set to either true or false, and pull.ff not set to
> >> only), which is a bit more gentle than the current situtation" a
> >> good stopping point.  That state is already making ff-only the
> >> default for unconfigured users, or you can view it as shipping "git
> >> pull" in a shape that has the more dangerous half of its feature
> >> disabled to avoid hurting users.  So I am not sure why you keep
> >> saying you do not have --ff-only as the default.
> >
> > The warning doesn't make the pull fail, ff-only does.
>
> Then probably you are giving an error and a warning at a wrong
> place.
>
>  - When history fast-forwards, and the user hasn't chosen between
>    rebase or merge, there is no need to give any warning.  Just
>    succeed by fast-forwarding.

Yes. That's what my patch [1] in this series does.

Show 5 quoted lines

>  - When history does not fast-forward and the user hasn't chosen
>    between rebase or merge, whether pull.ff is set to "only" or not,
>    we should fail and the error message can instruct the user to
>    choose between rebase and merge; there is no "ff-only" option
>    that is useful in the situation.

Yes. *Eventually*, that's part 3.

Show 6 quoted lines

> And that essentially makes the "ff-only" mode the safe default that
> castrates one half of the feature (the more dangerous half) of "git
> pull".  Why do we make it more complicated than that by warning that
> the user must choose in the future?  They will see an error tell
> them that when they start pulling while on their own work, and I do
> not see a need to bother them before that point.

Because the amount of time users have seen the correct error message telling them about this upcoming change is *zero*.

No one is aware of this backwards-incompatible change that is upcoming, since the current warning message doesn't specify any such future change.

Moreover, no one has configured their pull.mode, because no one has seen the message that tells them to do so.

And no one has had time to try out the upcoming "pull.mode=ff-only". They haven't had time to find bugs on it, or weird interactions, or suggestions how to improve it. Because the code has not been deployed, and no warning has told them to tentatively enable that mode.

When the default for push.default was changed, it was a good thing that we gave users (and ourselves) a grace period to try it out before permanently flipping the switch. The transition went smoothly.

I think the progression of the warning is hard to see from all the patch series. I'm sending all my patches and I will explain in a timeline how it progresses until eventually we reach the desired end goal.

Cheers.

[1] https://lore.kernel.org/git/20201204061623.1170745-8-felipe.contreras@gmail.com/

-- 
Felipe Contreras