Hi, Sheplyakov Alexei wrote:

On Mon, Feb 19, 2007 at 03:45:03PM +0800, Jerome BENOIT wrote:

...

Let be more specific.

During some kind of computation I have to compute (multi-indexed) coefficients: 1] these coefficients are long to compute; 2] they are computed by sub set through (consuming) recurrence formulae; 3] these coefficients are meant to be used randomly and very often.

...

Clearly the remember machinery can be use to overcome point [1] and [3]: for GiNaC function this is transparent for the user.

Point [2] causes me some difficulty: I want to compute each subset only once.

...

Hence the idea to compute each subset through a GiNaC function (with the remember option) which return tables,

Why do you need _GiNaC_ function to compute coefficients? What's wrong with C++ function, e.g.

ex compute(const ex& data) { static exmap remember; exmap::iterator i = remember.find(data); if (i != remember.end()) return i->second;

ex result; // actually compute it, making use of compute() itself

remember.insert(exmap::value_type(data, result)); return result; }

...

Right now, my guess is that I have to introduce a super class with a exhashmap<SOMETHING> as (principal) member: I thought that there was a better GiNaC way to do so, but apparently not.

Unless I'm missing something fundamental, this approach seems to be unnecessary complicated.

I am more than agree: hence my question. I guess that I still too Maple oriented. Thanks you very much, Jerome