Re: Confusing error message for git switch -d

[Junio C Hamano <gitster@pobox.com>]

Anselm Schüler <mail@anselmschueler.com> writes:

> I wanted to get the code for CPython 3.12, so I tried running git
> switch -d 3.12, expecting 3.12 to be a tag (which it wasn’t, and it
> was a remote branch).
>
> Git produced this error message:
>
> fatal: '--detach' cannot be used with '-b/-B/--orphan'
>
> This is confusing since I didn’t pass any of these options.

True.  FWIW, this is shared between "checkout" and "switch', and I
agree that there is a huge room for improvement.

With (arguably overly aggressive for its own worth) the "--guess"
option being default, giving the name of a unique remote-tracking
branch (that is, "3.12" appears as a remote-tracking branch of only
one remote, among possibly multiple remotes you have defined for the
repository) would behave as if what you said, which is "3.12", is
equivalent to "-b 3.12 --track origin/3.12" (where the "origin"
thing is the "only one" remote for which 3.12 exists as a
remote-tracking branch).  So

    $ git switch --detach 3.12

behaves as if you said

    $ git switch --detach -b 3.12 --track origin/3.12

and the error message is given.

I am merely saying that the mechanism from which this error message
is given can be _explained_.  I do not think it is well designed, as
it does not give a great end-user experience.

Disabling checkout.guess configuration would force you into the
habit of being a bit more explicit.  Even with checkout.guess
configured to false, you can still say

    $ git switch --detach origin/3.12

and because this does not blindly translate to -b/--track, you would
not see such an error.

A more appropriate design for "--guess" would be to notice the
presence of "--detach" and act differently, i.e. not rewriting
blindly 3.12 to "-b 3.12 -t origin/3.12" unconditionally, instead
rewrite "--detacu 3.12" to "--detach origin/3.12" if "--detach" is
there.

Patches welcome ;-)