Re: Strong typing, dynamical languages, what to do?

[Noel Welsh <noelwelsh@yahoo.com>]

--- Jens_Axel_Søgaard <js@vgt-gym.dk> wrote:
> The paper have a very clear analysis; and it is very
> easy to follow the
> argumentation.
> This conclusion surprised me the first time I saw
> it:
> 
>   In fact, the overhead associated with allocating
> and collecting
>   cells from the heap can be reduced to less than
> one instruction per cell
> by
>   increasing the size of physical memory.
> 
> That's hard to beat manually.

And yet people have commented on the poor performance
of SML/NJ and one of SMl/NJ's distinguishing features
is heap allocation of everything (including call
frames).  Maybe memory isn't yet big enough.

(Of course my argument is based on hearsay.  It's a
pity SML/NJ can't run in a stack allocating as well as
a heap allocating mode; then we'd have empirical
evidence for the superiority of one mode over another
on current computers which would go a long way to
clearing up this argument.)

Noel

__________________________________________________
Do You Yahoo!?
Yahoo! Movies - coverage of the 74th Academy Awards®
http://movies.yahoo.com/