Re: Q: unboxed values and polymorphism

[Hannes Haug <Hannes.Haug@Student.Uni-Tuebingen.de>]

>>>>> "Robert" == Robert Dewar <dewar@cs.nyu.edu> writes:

    Robert> Hannes says "But will this give me unboxed integers and
    Robert> type information on the stack ?  I'd try a private type
    Robert> that's actually an integer. I'd use half of this integers
    Robert> (-2**30 ... 2**30-1) for my fixnums. Values outside this
    Robert> range would be indices in arrays of bignums or list
    Robert> cells. So I'd have unboxed integers and type information
    Robert> on the stack."

    Robert> You are encoding at a very low level, appropriate for C
    Robert> (since it is the only way to do things), but totally
    Robert> inappropriate for Ada. You need a variant record with a
    Robert> discriinant to indicate whether you have fixnums or
    Robert> indices. The discriminant would be a single bit.

Absolutely not. On the stack I have words and not records. I work at a
low level because it is necessary. I really don't want to have a explicit
bit. I don't want to wrap/unwrap my data. Otherwise I would choose the low
bits as tag bits. This has nothing to do with C or Ada. It has to do with
costs of type checks, tagging, untagging and arithmetic on tagged integers.
If you keep telling people that this is evil they will never implement lisp,
prolog, ... in Ada.

 -hannes