Representation of VM programs---applicable smobs?

[Mikael Djurfeldt <mdj at mdj dot nada dot kth dot se>]

Keisuke Nishida <kxn30@po.cwru.edu> writes:

> Guile doesn't understand my VM...

I think we should find a new representation of your programs.

I think they should look like a primitive procedure to the Scheme
user.

If you didn't need special treatment of marking, I would have
suggested a compiled closure.

One representation which probably would work is as a struct entity (a
applicable struct).  You would then have to compute a struct layout to
select which words should be GC protected, instead of computing this
at GC time as you do now.

I wish we instead already had the new struct representation...

A third possibility is to add applicable smobs to Guile---quite an
interesting idea actually.  It seems that this has a niche of its own,
just like smobs have their niche (simple, extensible, low-level
representation).

Actually, several of the current uses of compiled closures would fit
much better as applicable smobs (lower overhead).  This holds, for
example, for standard-eval-closures.

What are people's opinion about adding applicable smobs to Guile?
(This implies adding a new entry for scm_tc7_smob in the application
 switches in SCM_CEVAL and SCM_DAPPLY.)

Before we have applicable smobs, a compromise is that you pack VM
program smobs inside a compiled closure (made using scm_makcclo).

You would store a VM apply as element 0 and the VM program smob as
element 1 in the closure vector.