From: Phillip Wood <phillip.wood123@gmail.com>
Date: Fri, 02 Oct 2026 14:11:31 GMT
Subject: Re: [PATCH v9 0/4] var: -z output, multiple variables, and broken-out idents
Message-ID: <33f24834-8e73-4230-bf0a-6809a85d9a86@gmail.com>
In-Reply-To: <20260926162048.30853-1-andrewpleeter@gmail.com>

Hi Andrew

On 26/09/2026 17:20, Andrew Pleeter wrote:
> This is a reroll of the single patch in v8, split up as Junio asked
> for, with the changes Phillip asked for on the list.
> 
> v8 was one commit doing four things at once. It is now four patches:
> 
>    1/4 converts the internal representation of a multi-valued variable
>        from a newline-joined string to a string_list, with no change in
>        output. Phillip suggested doing this as its own preparatory step.
> 
>    2/4 adds "-z", which is what forces the switch to parse_options(), so
>        that conversion lives here rather than in 1/4.
> 
>    3/4 teaches "git var" to take more than one variable.
> 
>    4/4 adds GIT_AUTHOR_NAME and friends.

These patches look good to me. I've left a comment on the tests in 
patches 3 and 4, but I'm not sure they are enough to warrant a re-roll.

Thanks for working on this, supporting multiple values and NUL 
termination are really useful improvements.

Phillip

> Changes since v8:
> 
>   * GIT_SIGNING_KEY is dropped. Phillip asked three times how it was
>     meant to be used, and once I looked properly the answer was that the
>     value cannot be interpreted without also reading gpg.format, and that
>     in the default configuration it is a committer ident rather than a
>     key at all. I would rather leave it out than define a variable I
>     cannot describe. Details are in my reply to him on this thread.
> 
>   * Asking for several variables no longer exits non-zero just because
>     one of them has no value; such a variable is left out of the output
>     and the rest are still shown. A non-zero status is now reserved for
>     real errors, such as naming a variable that does not exist, so
>     callers can detect those from the exit code. A single variable still
>     exits 1 when it has no value, as before. This is Phillip's
>     GIT_CONFIG_NOSYSTEM point.
> 
>   * GIT_CONFIG_GLOBAL is documented as a variable that can have more than
>     one value.
> 
>   * The commit messages are prose rather than a list of bullet points,
>     and no longer narrate how the patch was developed.
> 
> The tests use nul_to_q rather than running test_cmp over files
> containing NUL bytes, which Phillip pointed out in v6. Each patch builds
> and passes t0007 on its own.
> 
> One thing I did not do, and would like an opinion on: for a multi-valued
> variable in multi-variable mode, rather than emitting a trailing
> delimiter, each value is shown as its own "VARIABLE=value" entry, which
> matches "git var -l". Phillip suggested the trailing delimiter and I am
> happy to switch.
> 
> Andrew Pleeter (4):
>    var: represent multi-valued variables with a string_list
>    var: add "-z" output mode
>    var: accept more than one variable
>    var: add broken-out identity variables
> 
>   Documentation/git-var.adoc |  68 +++++++++--
>   builtin/var.c              | 242 +++++++++++++++++++++++++++++--------
>   t/t0007-git-var.sh         | 153 +++++++++++++++++++++++
>   3 files changed, 403 insertions(+), 60 deletions(-)
>