[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Scheme-reports] 6.2 numbers

Andy Wingo scripsit:

> The initial note about numerical computation being neglected in lisps is
> stale.


> And actually, relative to the R6RS, this report does indeed neglect
> important areas, like the behavior of negative zero in transcendental
> functions, and the various nan/infinity interactions.
> Perhaps this was the right choice for this report; I don't know.

There's already a ticket for this.

>   Suggestion: If you don't want to specify everything there is to do
>   with -0.0 et al, suggest that implementors produce systems that behave
>   in accordance to the IEEE recommendations, 

I think we will certainly do that.

>   follow the recommendations of the R6RS.  

We may do that, but I'd rather be self-contained.

> Also: R6RS changed the spelling of the cumbersome "inexact->exact" and
> "exact->inexact" to "exact" and "inexact", respectively.  This is a
> positive step.

These are in IEEE Scheme, so we'd need to provide them as well as their
replacements.  If you don't like them, it's easy to provide the R6RS
versions for yourself.

> 6.2.3 talks a lot about flonums, but does not (that I could see)
> actually refer to what that means.

I've added this sentence:

    Fixnums are integers with a limited and machine-dependent range;
    flonums are real numbers with a limited and machine-dependent
    range and precision.

> There are warnings about equality predicates for inexact numbers, but no
> warning about nan.

Part of the above ticket.

> It's cool that you adopted Taylor's division operators.  To the extent
> that all five sets, are useful, a sixth is useful as well, `centered/'
> et al.  

Ticket filed.

> Also there is a typo in (OPERATOR/) (p.28): an extra space before
> `values'.

It's a layout effect: there is only one space in the source.

> The various div-and-mod operators could use some more examples.  Modulo
> and remainder get half a page, whereas the ones that you are often
> interested in get no space.  A table, perhaps?  Perhaps Taylor should
> publish his draft somewhere in some referenceable form.

Ticket filed.

> R6RS extended `number->string' to take a third argument, "precision".
> Perhaps this was already brought up on the list but if not, it might be
> a good addition.

I think multiple precision is basically a leftover.  Of my 21 schemes, only
one actually represents it.  For that matter, the free Common Lisps other
than CLISP (which does pure software floating point) don't bother with it

No saves, Antonio, loke es morirse en su lingua. Es komo            John Cowan
kedarse soliko en el silensyo kada dya ke Dyo da, komo          cowan@x
ser sikileoso sin saver porke.                      http://www.ccil.org/~cowan
                        --Marcel Cohen, 1985

Scheme-reports mailing list