Re: [PATCH] builtin/config.c: prefer `--type=bool` over `--bool`, etc.

[Taylor Blau <me@ttaylorr.com>]

On Thu, Mar 29, 2018 at 06:11:22PM -0400, Jeff King wrote:
> > +Valid `[type]`'s include:
> > ++
> > +- 'bool': canonicalize  values as either "true" or "false".
> > +- 'int': canonicalize  values as simple decimla numbers. An optional suffix of
> > +  'k', 'm', or 'g' will cause the value to be multiplied by 1024, 1048576, or
> > +  1073741824 prior to output.
> > +- 'bool-or-int': canonicalize according to either 'bool' or 'int', as described
> > +  above.
> > +- 'path': canonicalize by adding a leading `~` to the value of `$HOME` and
> > +  `~user` to the home directory for the specified user. This specifier has no
> > +  effect when setting the value (but you can use `git config section.variable
> > +  ~/` from the command line to let your shell do the expansion.)
> > +- 'expiry-date': canonicalize by converting from a fixed or relative ate-string
> > +  to a timestamp. This specifier has no effect when setting the value.
> > ++
>
> Yay. It's nice to have this in only one place now.

Thanks! Agreed :-).

> s/ate-string/d&/ :)

Ack.

My excuse for this is that I have started using iTerm2.app with Vim in
my terminal using the new graphics acceleration options, and have had
trouble getting Vim to render the underline for misspelled words
consistently.

> > +static int type_name_to_specifier(char *name)
> > +{
> > +	if (!(strcmp(name, "bool")))
> > +		return TYPE_BOOL;
>
> We'd usually drop the extra level of parentheses, and just write:
>
>   if (!strcmp(name, "bool"))

Sounds good; I have adjusted this in the appropriate location in v2.

> > @@ -601,6 +618,14 @@ int cmd_config(int argc, const char **argv, const char *prefix)
> >  		usage_with_options(builtin_config_usage, builtin_config_options);
> >  	}
> >
> > +	if (type) {
> > +		if (types != 0) {
> > +			error("usage of --type is ambiguous");
> > +			usage_with_options(builtin_config_usage, builtin_config_options);
> > +		}
> > +		types = type_name_to_specifier(type);
> > +	}
>
> This error message left me scratching my head for a minute. Ambiguous
> how? I think this is covering the case of:
>
>   git config --int --type=bool
>
> So maybe "--type cannot be used with other type options" or something?
>
> Let's take a step back, though. As part of this, should we convert the
> parsing of type options to last-one-wins? The fact that they are all
> OPT_BIT() is quite silly, since you cannot have more than one bit set.
> So if you do:
>
>   git config --int --bool
>
> you get an error. Whereas normal behavior for most options would be for
> --bool to override --int. And that is what happens with:
>
>   git config --type=int --type=bool
>
> I don't think there are any backwards compatibility issues to deal with
> here; we'd be changing a case which is now always an error.

Agreed.

> And then after that, you truly can make (and document, if we want) that
> "--int" is a true synonym for "--type=int".

Great point; for me this is the primary motivating factor for making
this change. In addition to simplifying our work when we check:

  if (types != 0 && type_str) { ... }

it would be nice not to have to compare the two in order to ensure that
they are the same, continuing on if so, and causing an error if not.

> I think it would be pretty simple. One of:
>
>   - convert OPT_BIT("bool") into OPT_CALLBACK("bool") and just assign
>     "bool" to the "type" string, which will then later get parsed into
>     TYPE_BOOL.
>
> or
>
>   - convert OPT_BIT("bool") into OPT_SET_INT("bool") to set TYPE_BOOL
>
>     directly. Convert OPT_STRING("type") into OPT_CALLBACK(), and have
>     it assign the result of type_name_to_specifier() directly.
>
> I'd probably do the latter, but would be fine with either (and I'd make
> the OPT_SET_INT thing its own preparatory patch).

I adopted your advice and used the later, converting each `OPT_BIT` into
an `OPT_SET_INT`, and adding a callback for `--type`, which does _not_
complain if `type` is already non-zero.

> If you really want to go all-out, I think the ACTION flags could use the
> same cleanup. We treat them as bitflags, and then issue an error when
> you set more than one, which is just silly.

Agreed, and I think that this is a good candidate for a future patch.
Thoughts? :-).


Thanks,
Taylor