Path: utzoo!attcan!uunet!mcvax!hp4nl!botter!star.cs.vu.nl!jos
From: jos@cs.vu.nl (Jos Warmer)
Newsgroups: comp.lang.eiffel
Subject: Re: Eiffel type system <133@eiffel.UUCP>
Keywords: Eiffel type system, double precision, C interface, composite objects, copy semantics
Message-ID: <2717@vlot.cs.vu.nl>
Date: 8 Jun 89 13:55:58 GMT
References: <132@eiffel.UUCP> <133@eiffel.UUCP> <154@eiffel.UUCP>
Reply-To: jos@cs.vu.nl (Jos Warmer)
Organization: VU Informatica, Amsterdam
Lines: 63

In article <154@eiffel.UUCP> bertrand@eiffel.UUCP (Bertrand Meyer) writes:
>
>     As  an  aside,  note  that  release  2.2  provides
>     further  support  for  more  powerful  interaction
>     between   Eiffel   and   other   languages.     In
>     particular,  a  simple syntactical extension makes
>     it possible to  pass  the  address  of  an  Eiffel
>     routine  to  an external routine.  The notation is
>     @f, where f is a routine of the  enclosing  class;
>     such   an  expression  is  only  valid  as  actual
>     argument to an external routine.
>
I guess/hope that this can be used as an efficient interface for calling
Eiffel routines from C. But I don't understand the meaning of this.

An Eiffel routine has meaning only when it is called on an Eiffel
object as in 'object.routine'.  An Eiffel routine can never be
called as a standalone function as C routines can.
Besides, because of the dynamic binding in Eiffel, you can never
be sure what implementation of a routine is called.
Even when the names of the routines are equal and they are called on an
object of the same entity type, they can be different implementations.

So, how can this address be used by the external routines ?
They (or hopefully the Eiffel runtime system) must ensure that the routine
will only be called on an Eiffel object of the correct (dynamic) type.

If the routine address is used by the external routine to call
it directly, this check cannot be performed by the Eiffel runtime system.
This means that the external routines can easily corrupt internal Eiffel
structures.

If the routine address cannot be used by the external routine to call
it directly, what use is it?

____ABOUT_THE_EFFICIENCY_ISSUE____

Some time ago, I mentioned to ISE that the way of calling Eiffel routines 
from C was very inefficient.  In version 2.1b this is accomplished by the
C routine `eif_rout', which is declared as follows:

    DATUM eif_rout (Objptr, r_name, arg1, ...
    OBJPTR Objptr;
    char *r_name;

`Objptr' is a pointer to the eiffel-object and `r_name' is the
name of the routine that must be performed on `Objptr'.

In the implementation of `eif_rout', the array containing the names of
all applyable routines for `Objptr' is searched for `r_name'.
If `r_name' is found, the corresponding routine is called.

This means that a number of string-comparisons will have to take place
for each function call.  I am not a fan for efficiency at each cost,
but when this interface must be used heavily, this is prohibitive.

                                 Jos Warmer
				 jos@cs.vu.nl
				 ...uunet!mcvax!cs.vu.nl!jos
-- 
                                 Jos Warmer
				 jos@cs.vu.nl
				 ...uunet!mcvax!cs.vu.nl!jos