Re: [PATCH] alias: pass --help through to shell alias

[Ian Wienand <iwienand@redhat.com>]

On Fri, May 24, 2024 at 05:34:55PM -0700, Junio C Hamano wrote:
> Ian Wienand <iwienand@redhat.com> writes:
> This is a safe behaviour because the underlying command may not be
> prepared to see "--help" and ignore it silently in the best case,
> e.g.
> 
>     sh -c 'false "$@" || :' - --help
> 
> that would confuse users by being totally silent, or 
> 
>     sh -c 'awk "$@" || :' - --help
> 
> that gives "awk: not an option: --help" (which is less useful than
> the report of alias),

The proposed patch does still print the alias expansion before trying
to pass the "--help" onto the command.  That was intention to be the
same as "-h" behaviour.  But I take your point the error message
doesn't help.

> or even worse yet, if the underlying command
> does not understand "--help" and considers it something different,
> who knows what havoc it would wreak.

I guess this means that someone was relying on git to filter out
--help such that it would never get to their command.  To me this
could probably be considered "undefined behaviour" that is probably
low risk to change, but I understand the aversion to it.

> For the above reason ...
> 
> > I would propose that "--help" to a shell alias is passed through to
> > the underlying command.  This way you can write aliases that act more
> > like the other git commands.
> 
> ... this is a dangerous thing to do unconditionally.
> 
> I wonder if we can come up with a notation to annotate aliases that
> do support the "--help" option that wouldn't have been used by
> mistake for existing aliases?

We could do something like make "!!" as an alias prefix also passes
this through.  It seems too niche to bother with though, TBH.

I could propose documenting the behaviour as-is, although that
probably moves it from undefined behaviour to "this will never
change".  I'll take advice on what you think :)

-i