Re: [PATCH v2 3/8] config: Use parseopt.

Felipe Contreras [off-list ref] writes:

+	if (HAS_MULTI_BITS(actions)) {
+		error("only one action at a time.");
+		usage_with_options(builtin_config_usage, builtin_config_options);

My initial reaction was:

	Can we easily say "--get and --getall are mutually incompatible"?

and knowing that it would take much more code, the second reaction was:

	Does the user know what we mean by "action"?

Since the answer to this is "Yes, the usage message comes from parseopt
and there is a clear categorization", I think the message is good enough.

What happens when the user says "config --get --get-colorbool user.name"?
Is it an error?  Is it diagnosed as an error?

It probably is easy to fix it by defining two bits of fake actions and do:

	if (get_color_slot)
        	actions |= ACTION_GET_COLOR;
	if (get_colorbool_slot)
        	actions |= ACTION_GET_COLORBOOL;

immediately before this HAS_MULTI_BITS check.

I know I suggested these to are type-like, but I realize that these two
are better categorized as actions tied to a specific type (color), as you
had in your earlier round.

+	if (actions == 0)
+		switch (argc) {
+		case 1: actions |= ACTION_GET; break;
+		case 2: actions |= ACTION_ADD; break;
+		case 3: actions |= ACTION_REPLACE_ALL; break;

Straight assignment, not ORing-in please.  It wastes a few seconds from
the reader wondering what other bits in the variable "actions" are used
for things other than ACTION_* (the answer is none).

Similarly, later conditions:

+	if (actions & ACTION_LIST) {

would read better if they used equality == checks.