Re: what should "git clean -n -f [-d] [-x] <pattern>" do?

From: Sergey Organov <sorganov@gmail.com>
Date: Jan 31, 2024, 13:04 UTC
Message-ID: <87plxhiri4.fsf@osv.gnss.ru>
In-Reply-To: <87il3enc1i.fsf@osv.gnss.ru>

Sergey Organov <sorganov@gmail.com> writes:

Show 53 quoted lines

> Junio C Hamano <gitster@pobox.com> writes:
>
>> Sergey Organov <sorganov@gmail.com> writes:
>>
>>> I'm still arguing in favor of fixing "-n", and I believe a fix is needed
>>> independently from decision about "-f -f".
>>
>> Even though I do not personally like it, I do not think "which
>> between do-it (f) and do-not-do-it (n) do you want to use?" is
>> broken.
>
> Well, you are right, but "-n" is not documented as "do-not-do-it" in the
> sense you use it here. 
>
>> It sometimes irritates me to find "git clean" (without "-f"
>> or "-n", and with clean.requireForce not disabled) complain, and I
>> personally think "git clean" when clean.requireForce is in effect
>> and no "-n" or "-f" were given should pretend as if "-n" were given.
>> I wish if it were "without -n or -f, we pretend as if -n were given,
>> possibly with a warning that says 'you need -f if you actually want
>> to carry out these operations'".
>
> Yep, then we'd not need "-n" that much, only if to cancel explicit "-f"
> (provided "-f -f" feature is removed.)
>
>>
>> But that is a separate usability issue.
>
> Yep, and that'd be very different design. 
>
>>
>> What I find broken is that giving one 'f' and one 'n' in different
>> order, i.e. "-f -n" and "-n -f", does not do what I expect.  If you
>> are choosing between do-it (f) and do-not-do-it (n), you ought to be
>> able to rely on the usual last-one-wins rule.  That I find broken.
>
> I fail to see where this expectation comes from, provided "-n" is not
> documented as anything opposed to "-f":
>
>        -n, --dry-run
>            Don’t actually remove anything, just show what would be done.
>
> This is typical convenient description of "dry run", and current "-n"
> implementation is rather close to the description, that I'd still
> rewrite to emphasize the primary goal of the --dry-run:
>
>
> With these descriptions, the last thing that I'd expect is "-n -f"
> removing my files.
>
> Overall, as I see it, we have buggy implementation of suitably
> documented "--dry-run" option, and the best course is to fix the
> bug, with no semantic changes to the option itself.

OTOH, to preserve current actual behavior as much as possible, we can probably first fix documentation like this:

        -n, --dry-run
            Show what would be done, and don’t actually remove anything.
            This sets 'clean.requireForce' to 'false' for the duration
            of this command execution.

that to me looks like a match for current observable behavior.

Then we can fix '-n' implementation exactly according to this updated specification, making '-n' really independent from '-f', yet keeping pure "git clean -n" as well as "git clean -f -n", and "git clean -n -f" backward compatible.

As a bonus, the above solution will also free our hands in [re]defining '-f -f' later, if needed.

WDYT?

Thanks, -- Sergey Organov