openssh unix dev - Jun 2021 - Reminder: Add ssh_config equivalents of -N, -n and -f

On Thu, 3 Jun 2021, Volker Diels-Grabsch wrote:
> Ping.
> 
> [ If there's anything I can do to reduce the barrier, such as posting
>   the pull request as patch files to the mailing list, or anything,
>   please let me know. ]
Looking at these now.

I'm not sure about the naming of NoShell, because it's really about not
starting any kind of remote session. E.g. the existing -N talks about
not executing a remote command. Talking about shells I think runs the
risk that people will assume it somehow skips the remote shell for
command execution. They would probably find out soon enough that they
were wrong, but IMO it's better to avoid it altogether.

So the options are to rename the option to something like NoSession,
find a way to reuse the existing RemoteCommand option (though that is
IMO a little ugly) or perhaps introduce a new "SessionType" option
that can subsume functionality of both the -N (SessionType=none) and
-s (SessionType=subsystem) flags.

What do you think?

-d

Dear Damien,

Thanks for your response!

Damien Miller wrote:
> I'm not sure about the naming of NoShell, because it's really about
not
> starting any kind of remote session.
I didn't really care about how to name this.  All I did was noting
that the flag is stored in a C variable "no_shell", which I just
translated to camel case.

So I assume that we should rename that C variable as well?
> So the options are to rename the option to something like NoSession,
> find a way to reuse the existing RemoteCommand option (though that is
> IMO a little ugly) or perhaps introduce a new "SessionType"
option
> that can subsume functionality of both the -N (SessionType=none) and
> -s (SessionType=subsystem) flags.
>
> What do you think?
I also don't think that reusing RemoteCommand is a good idea.  This
would add unnecessary entaglement of mostly-unrelated options, and as
such increase the risk of accidentally introducing bugs.

I'd prefer to rename NoShell to NoSession, and perhaps to do the same
for the internal variable names for consistency.  Then, we would have
a one-to-one match of command line and configuration file, which is
IMHO the least surprise for users already familiar with "-N" and
"-s".

However, I'm not opposed to introduce
"SessionType=none/subsystem", in
case you like me to implement it that way.

How should we proceed from here?  Which variant should I implement?


Best regards,
Volker

-- 
Volker Diels-Grabsch
----<<<((()))>>>----