Re: [PATCH v3 6/6] hook: allow out-of-repo 'git hook' invocations

On Tue, Aug 24 2021, Randall S. Becker wrote:

On August 24, 2021 4:12 PM, Ævar Arnfjörð Bjarmason wrote:

quoted

On Wed, Aug 18 2021, Emily Shaffer wrote:

quoted

Since hooks can now be supplied via the config, and a config can be
present without a gitdir via the global and system configs, we can
start to allow 'git hook run' to occur without a gitdir. This enables
us to do things like run sendemail-validate hooks when running 'git send-email'
from a nongit directory.

Sensible goal. Perhaps we should note in an earlier commit when config-based hooks are introduced something like:

To clarify the requirements here, if running without a gitdir (and
thus without a repository?) how will front-ends know what to supply?
Will this just be "some shell script" that runs?

Emily's also aiming to have "git hook run" running "some shell script",
but in this case we're talking about any git program that runs hooks,
but doesn't require a repo. I think the only one is git-send-email's
sendemail-validate hook.

quoted

   Even though we've added config-based hooks, they currently only work
   if we can find a .git directory, even though certain commands such
   as "git send-email" (or only that command?) can be run outside of a
   git directory. A subsequent commit will address that edge-case.

So we cannot assume anything about the repository, correct? Similar to running git version but not git status?

Right, with RUN_SETUP_GENTLY we may not have a repo at all, so like
commands such as "git bundle", "git config", "git ls-remote" etc.

quoted

quoted

[...]
Notes:
    For hookdir hooks, do we want to run them in nongit dir when core.hooksPath
    is set? For example, if someone set core.hooksPath in their global config and
    then ran 'git hook run sendemail-validate' in a nongit dir?

So this is complete consent to run outside of git? I wonder whether
there needs to be an attribute associated with the hook that enables
this edge capability. That way we can validate whether the hook should
be run or not (from front-end scripts).

If I understand you correctly you're pointing out that anyone with a
sendemail-validate script could previously have assumed a repo, but not
anymore. So e.g. someone who always has other config the hook requires
when they run it in repository might unexpectedly have that hook fail if
they naïvely set the config for that sendemail-validate hook via "git
config --global".

I think that's a good point, and one I hadn't really considered. I think
it's probably best to just document this edge case for the one (or few?)
hooks that have this caveat than to not support it. I.e. it seems useful
to have a sendemail-validate without a repo (just to parse/validate the
about-to-be-sent-email), that we needed a repo before was really only an
emergent effect.

Also, I think it's not new with this config-based hook mechanism, but
something that I introduced in back when I added core.hooksPath (but I
don't think I realized I was adding this edge case at the time).