Re: [PATCH] log --format: teach %C(auto,black) to paint it black only on terminals

From: Junio C Hamano <gitster@pobox.com>
Date: Dec 17, 2012, 19:34 UTC
Message-ID: <7v4njkmq07.fsf@alter.siamese.dyndns.org>
In-Reply-To: <20121217121354.GB21858@sigill.intra.peff.net>

Jeff King <peff@peff.net> writes:

Show 19 quoted lines

> On Mon, Dec 17, 2012 at 06:44:10PM +0700, Nguyen Thai Ngoc Duy wrote:
>
>> >                         if (!end)
>> >                                 return 0;
>> > -                       color_parse_mem(placeholder + 2,
>> > -                                       end - (placeholder + 2),
>> > +                       if (!memcmp(begin, "auto,", 5)) {
>> > +                               if (!want_color(-1))
>> > +                                       return end - placeholder + 1;
>> 
>> This want_color() checks color.ui and only when color.ui = auto, it
>> bothers to check if the output is tty. I think the document should say
>> that "auto," (or maybe another name because it's not really auto)
>> respects color.ui.
>
> Yeah, that should definitely be documented. I wonder if it should
> actually respect color.diff, which is what "log" usually uses (albeit
> mostly for the diff itself, we have always used it for the graph and for
> the "commit" header of each entry).

I actually do not like this patch very much. The original motive behind this "auto" thing was to relieve the script writers from the burden of having to write:

	if tty -s
        then
        	warn='%C(red)'
                reset='%C(reset)'
	else
        	warn= reset=
	fi
        fmt="${warn}WARNING: boo${reset} %s"

and instead let them write:

	fmt="%C(auto,red)WARNING: boo%C(auto,reset) %s"

but between the two, there is no $cmd.color configuration involved in the first place. I am not sure what $cmd.color configuration should take effect---perhaps for a "git frotz" script, we should allow the script writer to honor frotz.color=(auto,never,always) configuration, not just ui.color variable.

Also the patch as posted deliberately omits support to honor command line override --color=(auto,never,always), but it may be more natural to expect

    git show --format='%C(auto,red)%s%C(auto,reset)' --color=never

to defeat the "auto," part the script writer wrote.

Now, such a script would be run by its end users as

    $ git frotz --color=never

It has to do its own option parsing before running the underlying "git show" to decide if it passes "--color=never" from the command line for that to happen.

But at that point, we are back to the square one. Such a script would be doing something like:

	if cmdline_has_color_flag
        then
		use_color=... that flag ...
	elif git config --get-colorbool frotz.color
	then
		use_color=--color=always
	else
		use_color=--color=never
	fi

in its early part to decide $use_color, to be used in the call it makes to "git show" later on:

	git show --format="$fmt" $use_color

Adding the logic to decide if %C(...) should be added to $fmt no longer is an additional burden to the script writer, making the whole %C(auto,red) machinery of little use.

So...