Re: [PATCH] branch: issue "-l" deprecation warning after pager starts

On Wed, May 30, 2018 at 11:52:19AM +0900, Junio C Hamano wrote:

Jeff King [off-list ref] writes:

quoted

Right, what I meant by "gentler" is that we continue to perform the same
behavior as the old version, alongside the warning. It's arguable here
because running "git branch -l" has _always_ been wrong. It's just wrong
in a way that happens to do what the user wants. ;)
...

quoted

Anyways, if you think it mustn't turn into an error now and only in the
next stage, a suggestion follows in another thread.

I don't think "mustn't", but I have a slight preference for what I
posted, as I think it is a little friendlier during the transition (at
the risk of somebody missing the warning, but then step 2 turns it into
a hard error anyway, so they'll certainly find out then).

Well, we could keep treating '-l' given in contexts where we have
silently ignored the option and did "list" instead as before during
the transition, until the very end where it becomes an explicit
"list" command, no?  Then there is no need to even warn against '-l'
that is ignored because we are listing in the earliest step.  The
only usage that requires a warning then becomes '-l' used for its
original meaning to create a reflog, right?  That sounds gentler to
me.

So I think you're proposing:

  - step 0: warn about "-l" when it actually gets used, and otherwise
    continue ignoring

  - step 1: turn "-l" into "--list"

  - step 2: there is no step 2

I can't think of any reason that would work, and it lets us reclaim it
for "--list" sooner. I guess "when it gets used" is maybe not the right
criterion. We'd warn on:

  git branch -l foo

But we wouldn't on:

  git branch -d -l foo

That's clearly nonsense, but we're probably better off complaining. So I
guess the right rule is to warn when we are not in list-mode, and
otherwise quietly accept it.

That does mean that anybody who misses the deprecation warning may be
surprised when "branch -l foo" starts listing instead of creating a
branch with a reflog (whereas in the current 3-step plan, we have a
period in the middle where that's a hard error). That may be OK, though,
and is a natural consequence of getting to the end step sooner (even
with a 3-step plan, anybody who skips the versions in the middle _could_
be surprised).

-Peff