From: Andrew Pleeter <andrewpleeter@gmail.com>
Date: Sat, 26 Sep 2026 16:19:51 GMT
Subject: Re: [PATCH v8] var: support broken-down idents, signing key, multiple args, and -z
Message-ID: <20260926161951.30531-1-andrewpleeter@gmail.com>
In-Reply-To: <ddfd5232-b832-476c-8614-268f8e4bbe28@gmail.com>

Hi Phillip,

Sorry for the silence. I was replying on the GitGitGadget PR and did not
realise those comments never reach the list. That is also why v6 switched
to get_signing_key() without a word from me about your question.

You were right to keep asking about GIT_SIGNING_KEY. I am dropping it
from the series. get_signing_key() returns a string whose meaning depends
on gpg.format, and the variable gives the caller no way to tell which
kind it got:

  gpg.format=openpgp, user.signingkey unset
    $ git var GIT_SIGNING_KEY
    C O Mitter <committer@example.com>

  gpg.format=openpgp, user.signingkey=DEADBEEF
    $ git var GIT_SIGNING_KEY
    DEADBEEF

  gpg.format=ssh, user.signingkey=/path/to/key.pub
    $ git var GIT_SIGNING_KEY
    /path/to/key.pub

So no, the key alone does not tell a caller whether to use gpg or ssh;
they would have to read gpg.format as well. And in the default case the
value is a committer ident rather than a key, because that is what
get_signing_key() hands to gpg to look up. Better to drop it than ship a
GIT_SIGNING_KEY that sometimes isn't one. I will revisit it when I can
name a caller it serves.

On the exit code, your GIT_CONFIG_NOSYSTEM example convinced me. In v9,
asking for several variables omits the ones with no value and exits 0.
Non-zero is reserved for real errors such as an unknown variable name, so
callers can detect those from the exit code alone. A single variable
still exits 1 when it has no value.

The string_list conversion is split out as you suggested:

  1/4  var: represent multi-valued variables with a string_list
  2/4  var: add "-z" output mode
  3/4  var: accept more than one variable
  4/4  var: add broken-out identity variables

parse_options() sits in 2/4, since "-z" is what requires it.
GIT_CONFIG_GLOBAL is now documented as multi-valued, and the tests use
nul_to_q instead of test_cmp over files containing NUL.

One place I did not follow you: instead of a trailing delimiter for a
multi-valued variable when argc > 1, each value gets its own entry:

  $ git var GIT_AUTHOR_IDENT GIT_CONFIG_GLOBAL
  GIT_AUTHOR_IDENT=A U Thor <author@example.com> 1112912053 -0700
  GIT_CONFIG_GLOBAL=/home/me/.config/git/config
  GIT_CONFIG_GLOBAL=/home/me/.gitconfig

Same shape as "git var -l", so one parser handles both. Happy to switch
if you prefer the trailing delimiter. I should have raised it last time
instead of just changing it.

Thanks,
Andrew