[gambit-list] Re: Gambit-C 4.0 beta 16

Christian christian at pflanze.mine.nu
Thu Jan 5 06:48:58 EST 2006


At 12:03 Uhr -0800 04.01.2006, Alex Goldman wrote:
>Hi
>
>I've looked at the documentation, and I have some questions:
>
>What does Gambit do to reduce name clutter in bigger than toy
>programs? (...) Are there reader modifications that mimic Common 
>Lisp's packages?

It offers the undocumented ##namespace form. See examples/* in the 
source archive. I've written "chjmodule" which makes dealing with 
those easier, see my previous announcements on this mailing list.

##namespace is not a reader macro, but a compiler/interpreter 
directive. It can also be put into a local scope. Like:

(##namespace ("foo#"))
(define (display v)
    (##namespace ("" display newline))
    (display "v: ")
    (display v)
    (newline))
(##namespace ("" fun +))
(define (fun a b)
   (display (+ a b)))

; > (compile-file "foo.scm")
; #t
; > (load "foo")
; "/home/chris/schemedevelopment/gambit/mod/foo.o2"
; > (fun 1 2)
; v: 3

>Incidentally, I tried to use the reference implementation of SRFI-82,
>but it failed with beta 15.

Do you mean srfi-83? What about André v. Tonder's implementation?

>If you do use gambit-generated code in your C program, can you also
>use pthreads?

From what I gather there is an example in the gambit source archive 
(examples/pthread/).

>Gambit allows file-level declarations. Is it possible to declare a
>specific variable to be fixnum or flonum, for example, or would that
>be useless anyway?

You can also declare fixnum or flonum in a lexical context. But I 
think not on single variables. From what I can tell, it's about how 
the generic operations like + - expt etc. are specialized, not about 
the variables. You can always use unsafe lowlevel operations on 
particular variables:

(define (foo a b) ; a: fixnum, b: flonum
   (if (and (##fixnum? a)
            (##flonum? b))
       (##flonum.+ (##flonum.<-fixnum a) b)
       (error "invalid operands")))

>Is it possible to UNload a shared library?

((Assuming that you are talking about compiled gambit object files:) 
Probably not since it would require support from the gc ("when is it 
safe to unload code?"). Of course I'm not authoritative. I've been 
wondering about this too, for long running processes, but then: it 
shouldn't be a problem anyway, since the files are memory-mapped (on 
linux anyway) and thus will barely take any real memory once they are 
not used anymore. And how long would it take to fill up a 32 bit 
virtual address space with object files? Like if you're loading 100 
shared objects of 200kb each per day, it will take about 100 days.)

Christian.



More information about the Gambit-list mailing list