Mathemagix

[Joris van der Hoeven <address@concealed>]

On Fri, Oct 19, 2007 at 02:58:36PM +0200, Grégoire Lecerf wrote:
> That would be better to agree on a uniform and clean solution :-)
> Julien, could you try the attached file in replacement without any other 
> previously discussed modification?
> Joris, Bernard, does this solution seem ok?

Not sure that this will work in all circumstances,
besides the fact that it prevents from clean destruction.

The best fix remains not to use any global variables of
type generic in basix. Of course, replacing generic by string
for the MMX_* symbols does induce a slight performance penalty.
But maybe we should go for safety and portability first.

[...]
> TMPL_DEF
> class mixed_symbol_rep REP_STRUCT {
> public: // FIXME: should become private
>   static table<Mixed_symbol_rep*,C,eq_op>& mixed_symbol_entries () {
>     static table<Mixed_symbol_rep*,C,eq_op>* t =
>       new table<Mixed_symbol_rep*,C,eq_op>((Mixed_symbol_rep*) NULL);
>     return *t;
>   }
[...]
>   inline Mixed_symbol_rep (const C& ref2):
>     ref (ref2), h (soft_hash (ref)), shared (true) {
>     static table<Mixed_symbol_rep*,C,eq_op>& t =
>       Mixed_symbol_rep::mixed_symbol_entries ();
>     t [ref]= this;
>   }