Re: [PATCH 2/3] submodule--helper: teach config subcommand --unset

[Junio C Hamano <gitster@pobox.com>]

Denton Liu <liu.denton@gmail.com> writes:

> This teaches submodule--helper config the --unset option, which removes
> the specified configuration key from the .gitmodule file.
>
> Signed-off-by: Denton Liu <liu.denton@gmail.com>
> ---
>  builtin/submodule--helper.c | 15 +++++++++++----
>  t/t7411-submodule-config.sh |  9 +++++++++
>  2 files changed, 20 insertions(+), 4 deletions(-)
>
> diff --git a/builtin/submodule--helper.c b/builtin/submodule--helper.c
> index 0e140f176c..336e4429e6 100644
> --- a/builtin/submodule--helper.c
> +++ b/builtin/submodule--helper.c
> @@ -2149,15 +2149,21 @@ static int module_config(int argc, const char **argv, const char *prefix)
>  	enum {
>  		CHECK_WRITEABLE = 1
>  	} command = 0;
> +	enum {
> +		DO_UNSET = 1
> +	} unset = 0;
>  
>  	struct option module_config_options[] = {
>  		OPT_CMDMODE(0, "check-writeable", &command,
>  			    N_("check if it is safe to write to the .gitmodules file"),
>  			    CHECK_WRITEABLE),
> +		OPT_CMDMODE(0, "unset", &unset,
> +			    N_("unset the config in the .gitmodules file"),
> +			    DO_UNSET),
>  		OPT_END()
>  	};

The way this patch uses OPT_CMDMODE() is wrong.

The situation in which CMDMODE is meant to be used is that there are
multiple (that's two or more) choices the end user can make, and the
end user can choose only one of them at a time, iow, giving more
than one is an error.  

In such a case, the programmer would

 - prepare a single variable to store the single choice the end user
   makes the choice in and initialize it to zero.

 - have one OPT_CMDMODE() element for each valid choice, all
   pointing at the same variable, but with different value that is
   not zero.

The parse_options() call would then make sure that the variable is
set to non-zero value only once, to detect conflicting command modes
given from the command line.  The program then can read from the
single variable to see if the end user made any choice (or left it
0).

An example of typical and good use of OPT_CMDMODE() mechanism can be
seen in builtin/tag.c; the 'l(ist)', 'd(elete)' and 'v(erify)' are
the distinct operating modes of the "git tag" command (i.e. you do
not delete a tag while listing them), so the &cmdmode variable is
used to ensure only one of them (or none of them) is given.

The existing use of OPT_CMDMODE we see in this function anticipates
that there would be new operating modes added other than the
check-writable mode, so if you are making a new command mode that
cannot be used with the existing check-writable, then the right way
to use the OPT_CMDMODE() is to add to the command enum a new
non-zero value distinct from CHECK_WRITABLE, without introducing a
new "unset" variable.

If this --unset thing is truly independent from --check-writable,
i.e. if all four possible combinations of having and not having
these two options are valid, then you would not want to muck with
the &command variable, but in that case, wouldn't it make more sense
for the new --unset thing to simply be using OPT_BOOL()?  After all,
it is not like you are planning to add new oprating modes in the
"unset" family that is not "--unset" in the future, no?