Re: [PATCH] branch: allow "-" as a short-hand for "previous branch"

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

Johannes Schindelin <Johannes.Schindelin@gmx.de> writes:

> @@ -1420,6 +1420,12 @@ static int interpret_nth_prior_checkout(struct repository *r,
>  	const char *brace;
>  	char *num_end;
>
> +	if (namelen == 1 && *name == '-') {
> +		brace = name;
> +		nth = 1;
> +		goto find_nth_checkout;
> +	}
> +
>  	if (namelen < 4)
>  		return -1;
>  	if (name[0] != '@' || name[1] != '{' || name[2] != '-')

If a solution along this line works, it would be far cleaner design
than the various hacks we have done in the past, noticing "-" and
replacing with "@{-1}".  For one thing, we wouldn't be receiving a
"-" from the end user on the command line and in response say @{-1}
does not make sense in the context in an error message.  That alone
makes the above approach to deal with it at the lowest level quite
attractive.

In the list archive, however, you may be able to find a few past
discussions on why this is not a good idea (some of which I may no
longer agree with).  One thing that still worries me a bit is that
we often disambiguate the command line arguments by seeing "is this
(still) a rev, or is this a file, or can it be interpreted as both?"
and "-" is not judged to be a "rev", IIRC.

Luckily, not many commands we have take "-" as if it were a file and
make it read from the standard input stream, but if there were (or
if we were to add a command to behave like so), treating "-" to mean
the same thing as "@{-1}" everywhere may require the "does this look
like a rev?"  heuristics (which is used by the "earlier ones must be
rev and not file, later ones must be file and cannot be interpreted
as rev, for you to omit '--' from the command line" logic) to be
taught that a lone "-" can be a rev.

So it is quite a lot of thing that the new code needs to get right
before getting there.