Re: [PATCH v2 0/3] worktree: teach add to accept --reason with --lock

From: Junio C Hamano <gitster@pobox.com>
To: Eric Sunshine <sunshine@sunshineco.com>
Cc: Stephen Manz via GitGitGadget <gitgitgadget@gmail.com>,
	git@vger.kernel.org, Stephen Manz <smanz@alum.mit.edu>
Subject: Re: [PATCH v2 0/3] worktree: teach add to accept --reason with --lock
Date: Fri, 09 Jul 2021 09:03:38 -0700	[thread overview]
Message-ID: <xmqqv95j79ut.fsf@gitster.g> (raw)
In-Reply-To: <YOf+tANuox4LRD7d@flurp.local> (Eric Sunshine's message of "Fri, 9 Jul 2021 03:45:56 -0400")

Eric Sunshine <sunshine@sunshineco.com> writes:

> The reason I suggested re-purposing `add_opts.keep_locked` is to avoid
> polluting that structure with members having overlapping meaning, thus
> reducing the confusion-factor for clients of that structure (assuming
> that a tri-state `keep_locked` is indeed less confusing). That doesn't
> preclude adding a new variable or two local to the `add()` function to
> facilitate keeping `add_opts` pure. For instance, it might be as
> simple as the below patch.

True.  It is less trivial to construct the command line option
parser so that --reason=<why> and --lock can be given in any order
(e.g. they no longer can be a simple OPT_BOOL and OPT_STRING that
can be given independently but needs some postprocessing like your
patch did), but it is not rocket science and keeping add_opts struct
leaner will reduce the risk of runtime confusion, I would think, but
at the same time, the room for runtime confusion would probably be
minor to begin with---so I am fine, if the coder cannot cleanly
write the option parser to do so, with the code as posted.