Re: [PATCH v2 0/8] Specify hooks via configs

From: Adrian Ratiu <adrian.ratiu@collabora.com>
Date: Feb 21, 2026, 14:27 UTC
Message-ID: <87o6liw4s1.fsf@collabora.com>
In-Reply-To: <aZjuTSopOMvwR4hQ@fruit.crustytoothpaste.net>

On Fri, 20 Feb 2026, "brian m. carlson" <sandals@crustytoothpaste.net> wrote:

Show 36 quoted lines

> On 2026-02-18 at 22:23:44, Adrian Ratiu wrote:
>> Hello everyone,
>> 
>> v2 addresses all feedback received in v1.
>> 
>> This series adds a new feature: the ability to specify commands to run
>> for hook events via config entries (including shell commands).
>> 
>> So instead of dropping a shell script or a custom program in .git/hooks
>> you can now tell git via config files to run a program or shell script
>> (can be specified directly in the config) when you run hook "foo".
>> 
>> This also means you can setup global hooks to run in multiple repos via
>> global configs and there's an option to disable them if necessary.
>> 
>> For simplicity, because this series is becoming rather big, hooks are
>> still executed sequentially (.jobs == 1). Parallel execution is added
>> in another patch series.
>
> I'm interested in how you plan to make parallel execution work
> gracefully.
>
> We've already established that it's necessary to preserve stdout and
> stderr (including wiring them up to the TTY) so as to not break
> existing, widely deployed hooks, such as those in Git LFS.  That means
> that to get parallel execution where the hooks don't write over each
> other's output and fight for the terminal, you'd need to multiplex each
> one, including providing a PTY if the appropriate descriptor already has
> a terminal, such that the output is at the very least handled line by
> line and ideally batched into per-hook chunks.  Is that the plan, or do
> you plan to do it differently?
>
> I ask because situations where the hook output is not handled gracefully
> and hooks fight over output or where the existence of TTY on a file
> descriptor is not preserved will result in bug reports and broken tests
> for tools that use Git, which I think we'd all like to avoid.

Hi Brian,

Yes, this is all done already. Phillip Wood actually brought this TTY issue up in his review of the v1 parallel hooks series (many thanks). :)

> Of course, that's not in this series, so it may not even be written yet,
> but if it's not, then this is something to keep in mind for when it gets
> submitted.

Yes, that's a separate series [1] and I'm about to send v2 very soon.

Please review v2 directly when I send it because it will contain significant changes from v1. v1 is not worth reviewing at this point.

To give you a high level description:

All hooks continue to run sequentially (serialized) by default, just like before.

To run some of the hooks in parallel, I had to introduce an extension, because we need to break backwards compatibility by combining stdout and stderr and piping them through run-command's muxer, detached from the terminal. There is only 1 known hook requiring this extension (pre-push) but it's trivial to add more, if necessary.

Patrick's idea is to leave it up to the user to decide what to parallelize, because the user knows if their hooks are safe or not (eg if they write the same file or call the same program), or if it's ok the enable the extension or not.

Some hooks are known never to be safe to parallelize, in that case git will always enforce serial execution.

Please wait for v2 of that series, it's my top priority to get it out, Adrian

1: https://lore.kernel.org/git/20260204173328.1601807-1-adrian.ratiu@collabora.com/n