Re: [Chicken-users] a couple of questions about foreign functions

[Felix Winkelmann]

Am Wed, 21 Jan 2004 20:05:34 -0800 hat Valentyn Kamyshenko 
<address@hidden> geschrieben:

> Hello:
>
> a couple of questions:
>
> - in the chicken code, the Scheme strings seem to be never created on
> the stack. Are there particular reasons for this? I mean, is
> smth. wrong with the following piece of code:
>
> (foreign-lambda* scheme-object () #<<EOF
> char buf[13+sizeof(C_header)];
> ((C_SCHEME_BLOCK*)(void*)buf)->header=C_make_header(C_STRING_TYPE,13);
> memcpy(((C_SCHEME_BLOCK*)(void*)buf)->data, "Hello, world!", 13);
> return(buf);
> EOF
> )

When this function returns, `buf' will point into a dead stack area,
so the data is likely to be corrupted after the return.

> - are there advantages/disadvantages of using foreign-lambda* instead
> of ##core#primitive function in terms of performance or they are more
> or less equivalent?

Good question. ##core#primitive has a little bit less overhead, here
some timing results on my 900Mzh Athlon:

#>
void foo(C_word c, C_word self, C_word k) { C_kontinue(k, 
C_SCHEME_UNDEFINED); }
<#

(define foo (##core#primitive "foo"))

(define bar
  (foreign-lambda* void () "") )

(define baz
  (foreign-callback-lambda* void () "") )

(define (repeat n thunk)
  (do ([i n (sub1 i)])
      ((zero? i))
    (thunk) ) )

(print "foo:")
(time (repeat 10000000 foo))

(print "bar:")
(time (repeat 10000000 bar))

(print "baz:")
(time (repeat 10000000 baz))

Without optimizations:

foo:
   4.316 seconds elapsed
       0 seconds in (major) GC
       0 mutations
    9871 minor GCs
       0 major GCs
bar:
   4.787 seconds elapsed
       0 seconds in (major) GC
       0 mutations
    8538 minor GCs
       1 major GCs
baz:
  14.611 seconds elapsed
   1.882 seconds in (major) GC
10000000 mutations
    1920 minor GCs
    1814 major GCs

With -O2 -f:

foo:
   0.881 seconds elapsed
       0 seconds in (major) GC
       0 mutations
    2401 minor GCs
       0 major GCs
bar:
   1.051 seconds elapsed
       0 seconds in (major) GC
       0 mutations
    2536 minor GCs
       0 major GCs
baz:
  10.305 seconds elapsed
    1.89 seconds in (major) GC
10000000 mutations
    3889 minor GCs
    1813 major GCs

> - can I call error routines from C code (that is, inside
> foreign-lambda*, for example)?

If you mean error-routines that eventually invoke Scheme, then
you should use foreign-callback-lambda*. If you handle the
error in C (for example to display a message and exit), then
I see no problem.


cheers,
felix