Re: server stuff

[Stefan Monnier]

> > I agree that $TMP or $TMPDIR should be obeyed in the same way
> > as temporary-file-directory.
> 
> Actually, on consideration, that's wrong.  TMPDIR might not point to a
> directory with the right semantics (local, sticky).  ssh-agent, for
> instance, always uses /tmp.

I think the stickiness is a "nonissue" since TMPDIR should either
be sticky or writable only by the user (otherwise, you get many more
holes anyway).
I don't know if the "local"ness is an issue.

If we don't obey TMP or TMPDIR, we should still allow the
user some way to specify a different socket location so she
can have several active emacs servers at the same time.

> > I believe that the creation of such sockets is atomic similarly
> > to OCREAT (at least it failed when I tried it with a pre-existing socket),
> > so I think it's safe.  But I must admit not knowing for sure.
> 
> That's not the issue.  I think you're supposed to create the socket in
> a protected directory, which is what ssh and gnuserv do, at least.

Indeed, I had no idea that the access rights of sockets might be ignored
on some (non-negligible) systems.

> I'm also unsure, but you'll get savaged on bugtraq if you make a
> release which gets it wrong...

;-)

> > You can pass a :version argument to define-minor-mode instead of
> > calling (custom-add-version 'server-mode "21.4").
> 
> That's not documented and it doesn't seem to work.

Omygod, I never finished that work!
Thanks for pointing it out.


        Stefan