When to use SCM_DEFINE vs SCM_GPROC vs SCM_PRIMITIVE_GENERIC?

When to use SCM_DEFINE vs SCM_GPROC vs SCM_PRIMITIVE_GENERIC?

[Mark H Weaver]

I am writing C versions of R6RS div/mod/div0/mod0, as well as
div-and-mod and div0-and-mod0, for numbers.c.

I notice that numbers.c uses three different macros to define public
procedures: SCM_DEFINE, SCM_GPROC, and SCM_PRIMITIVE_GENERIC.  I don't
see an obvious logic to the choices of which macro to use.  For
example, `exact->inexact' uses SCM_GPROC whereas `inexact->exact' uses
SCM_DEFINE.

One obvious disadvantage to SCM_GPROC is that it doesn't accept a
documentation string.

Can someone please enlighten me about the relevant factors to decide?
What macro do you think I should use for the functions I'm writing?
Note that `div-and-mod' and `div0-and-mod0' will return two values.

One more question: do you think I should give div/mod/div0/mod0 their
short R6RS names by default?

    Thanks,
      Mark

Re: When to use SCM_DEFINE vs SCM_GPROC vs SCM_PRIMITIVE_GENERIC?

[Andy Wingo]

Hi Mark,

On Fri 28 Jan 2011 01:47, Mark H Weaver <mhw@netris.org> writes:

> I notice that numbers.c uses three different macros to define public
> procedures: SCM_DEFINE, SCM_GPROC, and SCM_PRIMITIVE_GENERIC.  I don't
> see an obvious logic to the choices of which macro to use.  For
> example, `exact->inexact' uses SCM_GPROC whereas `inexact->exact' uses
> SCM_DEFINE.

Yeah, I have no idea why exact->inexact would be extensible and
inexact->exact is not.  (And perhaps we should deprecate those functions
in favor of the R6RS spelling, `inexact' and `exact'.  Another topic.)
And I don't recall the practical difference between GPROC and
PRIMITIVE_GENERIC.

> What macro do you think I should use for the functions I'm writing?
> Note that `div-and-mod' and `div0-and-mod0' will return two values.

For now, SCM_DEFINE.  We can change that later to PRIMITIVE_GENERIC if
needed.

> One more question: do you think I should give div/mod/div0/mod0 their
> short R6RS names by default?

Yes, I think.  At least I remember coming to that conclusion at one
point, before I had to move away from that work.  We might want opcodes
for some of these, also.

Andy
-- 
http://wingolog.org/

Re: When to use SCM_DEFINE vs SCM_GPROC vs SCM_PRIMITIVE_GENERIC?

[Mark H Weaver]

Andy Wingo <wingo@pobox.com> writes:
> And I don't recall the practical difference between GPROC and
> PRIMITIVE_GENERIC.

I found the answer.  According to a commit message by Mikael Djurfeldt
in 2003, all uses of SCM_GPROC should be converted to use
SCM_PRIMITIVE_GENERIC.

  http://blog.gmane.org/gmane.lisp.guile.cvs/month=20030301

I will do this for all extensible functions in numbers.c.  I will also
make some other numeric functions extensible.  For example, it occurs to
me if we make `inf?', `nan?', and `finite?' extensible, then users will
be able to easily extend them to handle non-real complex values, if
desired.

    Best,
     Mark

Re: When to use SCM_DEFINE vs SCM_GPROC vs SCM_PRIMITIVE_GENERIC?

[Andy Wingo]

On Sat 29 Jan 2011 16:39, Mark H Weaver <mhw@netris.org> writes:

> Andy Wingo <wingo@pobox.com> writes:
>> And I don't recall the practical difference between GPROC and
>> PRIMITIVE_GENERIC.
>
> I found the answer.  According to a commit message by Mikael Djurfeldt
> in 2003, all uses of SCM_GPROC should be converted to use
> SCM_PRIMITIVE_GENERIC.
>
>   http://blog.gmane.org/gmane.lisp.guile.cvs/month=20030301
>
> I will do this for all extensible functions in numbers.c.  I will also
> make some other numeric functions extensible.  For example, it occurs to
> me if we make `inf?', `nan?', and `finite?' extensible, then users will
> be able to easily extend them to handle non-real complex values, if
> desired.

Sounds great to me.  Thanks for doing the archeology!

Andy
-- 
http://wingolog.org/