Re: [PATCH] push: unset PARSE_OPT_OPTARG for --recurse-submodules

[Jeff King <peff@peff.net>]

On Mon, Apr 27, 2020 at 02:44:08AM -0400, Denton Liu wrote:

> When the usage for `git push` is shown, it includes the following
> lines
> 
> 	--recurse-submodules[=(check|on-demand|no)]
> 			      control recursive pushing of submodules
> 
> which seem to indicate that the argument for --recurse-submodules is
> optional. However, we cannot actually run that optiion without an
> argument:
> 
> 	$ git push --recurse-submodules
> 	fatal: recurse-submodules missing parameter
> 
> Unset PARSE_OPT_OPTARG so that it is clear that this option requires an
> argument. Since the parse-options machinery guarantees that an argument
> is present now, assume that `arg` is set in the else of
> option_parse_recurse_submodules().

Yeah, I think this is the right solution. It looks like it was broken
since the option was introduced in d2b17b3220 (push: Don't push a
repository with unpushed submodules, 2011-08-20).

I wondered if it was copied from another similar option in another
command, and if so whether that option had the same problem. But it
doesn't look like it. The other --recurse-submodules options are all
PARSE_OPT_OPTARG, but they actually do something useful when they are
not given an option (they turn it to "on").

I don't know enough about the "push" case to say whether it would be a
good idea for it to behave similarly, but certainly your patch is an
improvement until somebody decides to look into it.

> diff --git a/builtin/push.c b/builtin/push.c
> index 6dbf0f0bb7..90f071fcf2 100644
> --- a/builtin/push.c
> +++ b/builtin/push.c
> @@ -434,10 +434,8 @@ static int option_parse_recurse_submodules(const struct option *opt,
>  
>  	if (unset)
>  		*recurse_submodules = RECURSE_SUBMODULES_OFF;
> -	else if (arg)
> -		*recurse_submodules = parse_push_recurse_submodules_arg(opt->long_name, arg);
>  	else
> -		die("%s missing parameter", opt->long_name);
> +		*recurse_submodules = parse_push_recurse_submodules_arg(opt->long_name, arg);

What a lousy diff. It would be much easier to read if we chose to
replace the "else" and keep the big complicated line intact. :)
Both Myers and --histogram give the diff above, but --patience gives:

diff --git a/builtin/push.c b/builtin/push.c
index 6dbf0f0bb7..ac6cc07c8c 100644
--- a/builtin/push.c
+++ b/builtin/push.c
@@ -434,10 +434,8 @@ static int option_parse_recurse_submodules(const struct option *opt,
 
 	if (unset)
 		*recurse_submodules = RECURSE_SUBMODULES_OFF;
-	else if (arg)
+	else
 		*recurse_submodules = parse_push_recurse_submodules_arg(opt->long_name, arg);
-	else
-		die("%s missing parameter", opt->long_name);
 
 	return 0;
 }

Obviously not a complaint about your patch. I'm always just curious to
see cases where the various diff implementations do better or worse than
each other.

> @@ -554,7 +552,7 @@ int cmd_push(int argc, const char **argv, const char *prefix)
>  		  PARSE_OPT_OPTARG | PARSE_OPT_LITERAL_ARGHELP, parseopt_push_cas_option },
>  		{ OPTION_CALLBACK, 0, "recurse-submodules", &recurse_submodules, "(check|on-demand|no)",
>  			N_("control recursive pushing of submodules"),
> -			PARSE_OPT_OPTARG, option_parse_recurse_submodules },
> +			0, option_parse_recurse_submodules },

This could collapse down to OPT_CALLBACK() now, though I don't think
it's a big deal either way.

-Peff