On Mon, Nov 21, 2022 at 02:33:06PM +0900, Junio C Hamano wrote:
> Thierry Reding <thierry.reding@gmail.com> writes:
> 
> > From: Thierry Reding <treding@nvidia.com>
> >
> > The git-am --no-verify flag is analogous to the same flag passed to
> > git-commit. It bypasses the pre-applypatch and applypatch-msg hooks
> > if they are enabled.
> >
> > Signed-off-by: Thierry Reding <treding@nvidia.com>
> > ---
> > Changes in v2:
> > - add test to verify that the new option works
> 
> > +-n::
> > +--no-verify::
> > +	By default, the pre-applypatch and applypatch-msg hooks are run.
> > +	When any of `--no-verify` or `-n` is given, these are bypassed.
> > +	See also linkgit:githooks[5].
> 
> I think the goal of this topic is to allow bypassing the checks made
> by these two hooks (and possibly future ones that validate the input
> to "am"), and there are at least two possible implementations to
> achieve that goal.  You can still run the hook and ignore its
> failure exit, or you can skip running the hook and pretend as if
> hook succeeded.
> 
> As it is documented that applypatch-msg is allowed to edit the
> message file to normalize the message, between the two, running the
> hook (to allow the hook to automatically edit the message) but
> ignoring its failure would be a more intuitive approach to "bypass"
> the check.  If the option were called --no-hook or --bypass-hooks
> then it would be a different story, though.
> 
> >  	assert(state->msg);
> > -	ret = run_hooks_l("applypatch-msg", am_path(state, "final-commit"), NULL);
> > +
> > +	if (!state->no_verify)
> > +		ret = run_hooks_l("applypatch-msg", am_path(state, "final-commit"), NULL);
> 
> And it seems that this took a less intuitive avenue of bypassing the
> hook completely.  I am not 100% convinced that this is the better
> choice (but I am not convinced it is the worse one, either).

Thinking a bit more about this, if we let applypatch-msg run but ignore
failures and continue on to commit the result, wouldn't that potentially
allow committing garbage? I'm thinking about cases where applypatch-msg
may attempt to normalize the message and fails badly, leaving a partial
commit message or none at all.

The primary use-case where I'd like to use this new option for git am is
when the pre-applypatch hook fails and that has less of the risks
associated with applypatch-msg, so perhaps --no-verify should only apply
to pre-applypatch?

> 
> > diff --git a/t/t4154-am-noverify.sh b/t/t4154-am-noverify.sh
> > new file mode 100755
> > index 000000000000..fbf45998243f
> > --- /dev/null
> > +++ b/t/t4154-am-noverify.sh
> > @@ -0,0 +1,60 @@
> > +#!/bin/sh
> > +
> 
> It is surprising, and I am not enthused to see, that this needs an
> entirely new script.
> 
> Don't we already have a script or two to test "am", among which the
> invocation of hooks is already tested?

I can move the tests to the corresponding sections in t/t4150-am.sh.

Thierry