Re: [PATCH] Ensure that SSH runs in non-interactive mode

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

Johannes Schindelin <Johannes.Schindelin@gmx.de> writes:

> How about this instead?
>
> -- snipsnap --
> diff --git a/connect.c b/connect.c
> index 574f42f..7e7f4d3 100644
> --- a/connect.c
> +++ b/connect.c
> @@ -603,7 +603,8 @@ struct child_process *git_connect(int fd[2], const char *url
>  
>  		*arg++ = ssh;
>  		if (port) {
> -			*arg++ = "-p";
> +			const char *opt = getenv("GIT_SSH_PORT_OPTION");
> +			*arg++ = opt ? opt : "-p";
>  			*arg++ = port;
>  		}
>  		*arg++ = host;

If you only care only about the ones we currently want to support, I do
not htink it makes any difference either way, but if we are shooting for
having a minimum-but-reasonable framework to make it easy to support other
ones that we haven't seen, it feels very much like an inadequate hack to
waste an envirnoment variable for such a narrow special case.  With this,
what you really mean is "Plink uses -P instead of -p", right?

I do not know if "plink" is used widely enough to be special cased, but if
so, I think we would better have an explicit support for it.  Will we add
GIT_SSH_FORBID_X11_FORWARDING_OPTION environment variable and friends,
too?

The extra environment would not help dealing with an implementation that
wants --port=90222 (i.e. not as two separate arguments but a single one),
for example.  You would need the extra wrapper support for that kind of
thing anyway.  That extra environment _solution_ will need to make an
assuption that any reasonable implementation would have an option string
to specify port which may not be "-p" and that is to be followed by a
separate argument that is a decimal port number, which probably is
reasonable for this particular "port" thing, but as a general design
principle I do not think it is a good direction to go.