Re: Generic Function Question

[email protected] Mon, 4 May 2020 08:20:35 +0200
Newsgroups gmane.lisp.scheme.bigloo
Organization Inria
Message-ID <[email protected]>
Hi Joe,

> Hello,
> While implementing srfi25 for bigloo, I experimented with using generic functions for the api and noticed that although you can have generics return non-bigloo types such as int64 the methods defined for that generic function always return boxed values. This is not a problem from a correctness perspective but it is a performance issue. For example in my scenario, the method was computing an int64 boxing it and then returning it to the generic function dispatcher which immediately unboxed it. Is there a way to remove this boxing/unboxing overhead?
> A very simple example demonstrating this is below.
> 
> (module example   (export         (generic test o::base i::long)  
>        (class base)       (class deriv::base))
>    (main main))
> 
> (define-generic (test::int64 o::base i::long))
> (define-method (test::int64 o::deriv i::long)    #s64:0)
> (define (main args)    (print "result: " (test (instantiate::deriv) 0))
Unfortunately, no. That's current not possible because the compiler always
box/tag values returned from closure, and methods attached to generic
functions are closures.

I agree that this is really unfortunate but as I said there is no
immediate solution. Do you absolutely need signed int64 for srfi25?  As
a temporary work around can you go along with simple "long" as there is
(almost) no cost for tagging/untagging them? If you are using these
values to index arrays, using a 61-bit wide number should be enough, no?

-- 
Manuel