Re: [PATCH 5/5] config: make 'git config list --type=<X>' work

[Patrick Steinhardt <ps@pks.im>]

On Tue, Feb 10, 2026 at 04:42:59AM +0000, Derrick Stolee via GitGitGadget wrote:
> From: Derrick Stolee <stolee@gmail.com>
> 
> Previously, the --type=<X> argument to 'git config list' was ignored and
> did nothing. Now, we add the use of format_config() to the
> show_all_config() function so each key-value pair is attempted to be
> parsed.
> 
> If there is an error in parsing, then the row is not output.

I was a bit surprised at first, but now that I think about it a bit more
I think this is sensible behaviour. If I ask for `git config list
--type=int`, then I don't want to see any non-int configuration. I
wouldn't even expect a warning, as the option essentially works like a
filter.

> This is a change in behavior! We are starting to respect an option that
> was previously ignored, leading to potential user confusion. This is
> probably still a good option, since the --type argument did not change
> behavior at all previously, so users can get the behavior they expect by
> removing the --type argument or adding the --no-type argument.

Yeah, I fully agree that this is a sensible change in behaviour. It is
obviously broken right now, so I would claim that this is simply a bug
fix.

> diff --git a/Documentation/git-config.adoc b/Documentation/git-config.adoc
> index ac3b536a15..5300dd4c51 100644
> --- a/Documentation/git-config.adoc
> +++ b/Documentation/git-config.adoc

The synopsis of `git config list` should also be amended.

> diff --git a/builtin/config.c b/builtin/config.c
> index e69b26af6a..c83514b4ff 100644
> --- a/builtin/config.c
> +++ b/builtin/config.c
> @@ -363,21 +363,12 @@ static int show_all_config(const char *key_, const char *value_,
>  {
>  	const struct config_display_options *opts = cb;
>  	const struct key_value_info *kvi = ctx->kvi;
> +	struct strbuf formatted = STRBUF_INIT;
>  
> -	if (opts->show_origin || opts->show_scope) {
> -		struct strbuf buf = STRBUF_INIT;
> -		if (opts->show_scope)
> -			show_config_scope(opts, kvi, &buf);
> -		if (opts->show_origin)
> -			show_config_origin(opts, kvi, &buf);
> -		/* Use fwrite as "buf" can contain \0's if "end_null" is set. */
> -		fwrite(buf.buf, 1, buf.len, stdout);
> -		strbuf_release(&buf);
> -	}
> -	if (!opts->omit_values && value_)
> -		printf("%s%c%s%c", key_, opts->delim, value_, opts->term);
> -	else
> -		printf("%s%c", key_, opts->term);
> +	if (format_config(opts, &formatted, key_, value_, kvi, 0) >= 0)
> +		fwrite(formatted.buf, 1, formatted.len, stdout);
> +
> +	strbuf_release(&formatted);
>  	return 0;
>  }
>  

I wonder whether there is a good argument to be made here that we should
keep the old logic in case no "--type=" parameter was given. In that
case, for example the following output would remain the same:

> diff --git a/t/t1300-config.sh b/t/t1300-config.sh
> index 9850fcd5b5..b5ce900126 100755
> --- a/t/t1300-config.sh
> +++ b/t/t1300-config.sh
> @@ -2459,9 +2459,10 @@ done
>  
>  cat >.git/config <<-\EOF &&
>  [section]
> -foo = true
> +foo = True
>  number = 10
>  big = 1M
> +path = ~/dir
>  EOF
>  
>  test_expect_success 'identical modern --type specifiers are allowed' '

I'm not really sure whether we want that though. I actually like that
this also leads to some code duplication, so maybe this is fine?

Patrick