Re: orthogonal cases of log --date option

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

Jeff King <peff@peff.net> writes:

> Yep, it is more invasive. But I consider it more maintainable in the
> long run.

Why do you think it is more confusing to ask "--date=local --date=iso"
than asking "--local-time --date=iso"?  If the patch under discussion were
not mine, I would have said that --date=local that flips the "lie about
timezone" bit and tells us to use the "default" format is a brilliant and
elegant solution.

I honestly do not see the point of what you are proposing to make
"selector" and "format" independent; unless you are shooting for
"--use-tz=Indian/Christmas --date=iso", that is.

The "--use-tz=zonename" might make some sense.  The required change would
look more like:

 (1) Introduce:

        const char *force_output_tz;

     that defaults to NULL;

 (2) Option parser for --use-tz=Indian/Christmas would store
     arg+9 to force_output_tz;

 (3) Option parser for --date=local would set the date format to
     "default", and store "localtime" to force_output_tz.  We discard
     DATE_LOCAL enum.

 (4) In show_date(), instead of 

	if (mode == DATE_LOCAL)
		tz = local_tzoffset(time);

     you'd do:

	if (forced_output_tz)
		tz = zonename_to_tzoffset(time, force_output_tz);

     zonename_to_tzoffset() would look up the zoneinfo database and
     compute the offset in our internal (hour*100+min) format.

But I do not think that is what you were proposing.