Re: [PATCH v2 2/3] push: Add support for pre-push hooks

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

Junio C Hamano <gitster@pobox.com> writes:

> At least the attached patch is necessary.

Sorry, but the last hunk (see below) is not.  It breaks the hook.

> In the longer term, we may want to discuss what should happen when
> the hook exited without even reading what we fed.  My gut feeling is
> that we can still trust its exit status (a hook that was badly coded
> so it wanted to read from us and use that information to decide but
> somehow died before fully reading from us is not likely to exit with
> zero status, so we wouldn't diagnosing breakage as a success), but
> there may be downsides for being that lax.
>
> If we decide we want to be lax, then the call site of this hook and
> the pre-receive hook (is there any other "take info from the
> standard input" hook?) need to be modified so that they ignore
> sigpipe, I think.
>
> There was a related discussion around this issue about a year ago.
>
> http://thread.gmane.org/gmane.comp.version-control.git/180346/focus=186291
> ...
>

> @@ -38,6 +40,7 @@ COMMIT2="$(git rev-parse HEAD)"
>  export COMMIT2
>  
>  write_script "$HOOK" <<'EOF'
> +cat >/dev/null
>  echo "$1" >actual
>  echo "$2" >>actual
>  cat >>actual

As this one wants to keep the incoming data to "actual", we do not
want the extra "cat" to slurp everything in.  Sorry for not being
careful.