Re: Guile numerical work and uniform arrays

[Valentin Kamyshenko <val@kamysh.materials.kiev.ua>]

>>>>> "Clark" == Clark McGrew <mcgrew@ale.physics.sunysb.edu> writes:

>>>>> "Valentin" == Valentin Kamyshenko <val@kamysh.materials.kiev.ua> writes:
>>>>> "Jim" == Jim Blandy <jimb@red-bean.com> writes:

Clark> I think the full complements of types would be useful.  I don't
Clark> use doubles or long doubles at all, but I would use short ints
Clark> and unsigned short ints.  I guess it depends on the particular
Clark> problem.
I agree, it's worth to have representations of all types in guile. So
far as I uderstood from previous discussion, there exists a problem ,
that there are too many types of uniform vectors to be represented
simultaneously in guile, and we have to choose what are the most
commonly used. Am I right?

Valentin> Finally, I'd like to propose to think about
Valentin> multi-dimensional arrays representation.

Clark> I think the current Guile multi-dim arrays can easily be added
Clark> to the discussion.  If my understanding is correct it keeps
Clark> extra information to describe the dimensions, and then a
Clark> pointer to a chunk of memory to hold the nD array.
My impression is that there is also an array of pointers to one-dim
D-arrays. Otherwise, I think, it would be impossible to work with
shared arrays. My simple experiments show, that if I created a
multi-dim array, the undelying chunk existed, and could be treated as
one-dim nD array. I'm not sure, however, that this is always
guaranteed. My suggestion was, thus: 1) to give user the possibility
to know that a given multi-dimensional array can be treated as
one-dimensional; 2) to have a standard procedure to get a pointer to
this chunk.

--
Best regards,
	Valentin.