Re: Problem with inefficiency and records with variant parts

[eachus@mitre-bedford.arpa (Robert I. Eachus)]

In article <SIDOU.92Oct5232754@litsun.epfl.ch> sidou@litsun.epfl.ch (Dominique 
Sidou) writes:

   We have a big problem with records variant an efficiency...
   we now need to do the following...

   v.s := 2;
   fill(v.c2);

   with: procedure fill (c : in out BigStruct2);

   We want to avoid casting v in the following way before
   the call to fill:

   declare
     bigstruct2var : BigStruct2;
   begin
     v := (2, bigstruct2var); -- very inefficient and useless in our case
     fill(v.c2);
   end;


   Any Comments, Ideas...???

   The problem is one of different levels of abstraction.  Procedure
fill expects objects of type BigStruct2, and you want to operate on on
objects of type t.  Two possible solution approaches:

   First approach is to wrap fill to return a value of type t:

   function Fill return T is 
     Temp: t(2);
   begin
     fill(Temp.c2)
     return Temp;
   end Fill;
   ...
   v := Fill;

   (You could make it a procedure if you like, in that case you have
to depend on your optimizer to be more sophisticated than in this case.)

   Second approach is to redefine fill to return a value of type
BigStruct2, to be a procedure with an in out parameter of type, or to
match the template above.  Any of these should work without depending
on the optimizer.

   You may have simplified things in this example to eliminate things
like bounds on the component types which are in fact unconstrained
array types.  In that case it is probably easier on the reader of the
code to have function calls with explicit size parameters, but poor
optimizers will work better with in out parameters.


--

					Robert I. Eachus

with STANDARD_DISCLAIMER;
use  STANDARD_DISCLAIMER;
function MESSAGE (TEXT: in CLEVER_IDEAS) return BETTER_IDEAS is...