Re: TLS implementation

On Mon, May 4, 2015 at 9:46 PM, Stephen J. Turnbull <[email protected]> wrote:
> Jerry,
>
> I have a number of questions about how the TLS implementation
> interacts with the build system and to some extent the Lisp engine.  I
> know you're very busy, so if you don't have time to do any review or
> answer my question, I'd appreciate a note saying so and I'll just
> proceed with caution.

A note for those who haven't heard: I just changed jobs last week.
While I am enjoying the new job, my commute time went up quite a bit,
which means my free time went down commensurately.  I will still
contribute as I am able to do so, but may be less responsive than
previously.

> (0) I think we ought to default this feature on.  I fuzzily remember
>     that I wanted some testing before doing that, but most of the
>     build issues should be worked out by now, and the "fill can
>     relocate strings" issue has been fixed with Aidan's implementation
>     of clear_passwd.

I think that's great.  Mats and I talked about doing some configure
magic to prevent using pre-1.0.0 versions of openssl.  There was a
patch floated on xemacs-beta if I remember correctly.  That should
probably go in, as the openssl support won't work with 0.9* versions.

> (1) I guess the reason for always building and linking tls.c is to
>     define 'tls-error.  Is that correct?

Uhhhhhh....  I don't remember this terribly well, but glancing at the
code, that seems likely.

> (2) If I configure and build --without-tls, I get a warning because
>     Qread_passwd is an unused variable.  I don't see any reason not to
>     DEFSYMBOL it unconditionally, it's just one slot in the obarray
>     and this is a pretty important function.

Okay, that's fine with me.

> (3) I'm very tempted to move the definitions of 'tls-error to
>     process.c (and maybe make it a sub-error of 'network-error) and
>     'read-password to general-slots.h.  Any reason not to do that?

No, that sounds like a good plan.

> (4) Once (3) is done, I believe that tls.c doesn't need to be compiled
>     or linked if configured --without-tls.  Any reason not to do the
>     usual Makefile dance for conditionally compiled modules that you
>     know of?

No, I can't think of any.  I like this idea.  Feel free to ask me to
review any proposed changes.  I may take a day or two do to the
review, but I'll make an effort to get to it.  Regards,
-- 
Jerry James
http://www.jamezone.org/