Re: Generics and parameter modes

[Tucker Taft <stt@averstar.com>]

Martin Dowie wrote:
> 
> i have an instance of a package similar to the generic listed below.
> when calling the 'size' routine the actual parameter is passed by copy
> and not by reference as i had expected. this is really annoying as the
> data object is a few hundred kbytes in size! is there anything in Ada95
> that allows/requires this?

Non-limited, non-tagged record types may be passed either by reference
or by copy.  On the other hand, I am quite surprised that a compiler
would go out of its way to pass such a large object by copy, since it
requires more work.  The only reasons I can imagine for doing this are:

   1) the compiler is trying to enable more generic code sharing
     (but that seems a bit far-fetched, since even in a shared
      generic it "knows" a_list is a record type); 
   2) there is some kind of alignment problem, where the actual parameter
     is not properly aligned for the code inside the generic (again
     a bit far-fetched);
   3) the actual parameter is marked volatile or atomic (see RM95 C.6(19));
   4) convention C_Pass_By_Copy applies to the type (I don't see such
       a pragma below);
   5) the compiler has a bug ;-).

Are you really sure that it is passing the parameter by copy?  I would
contact Rational and file a lousy-code-quality-but-not-quite-a-bug report.

> generic
>     type an_item is private;
>     maximum_number_of_items : positive;
> package generic_list is
>     subtype a_number_of_items is natural range 0 ..
> maximum_number_of_items;
>     subtype an_item_index is natural range 1 .. maximum_number_of_items;
> 
>     type a_list is private;
>     -- Lots of other stuff...
>     function size (list : in a_list) return a_number_of_items;
> private
>     type an_array_of_items is array (an_item_index) of an_item;
>     type a_list is record
>         current_size, current_item : a_number_of_items;
>         items : an_array_of_items;
>     end record;
> end generic_list;
> 
> the type used in the declaration is a variant record which it self
> contains variant fields within a number of the variant branches (approx.
> 300 bytes storage required).
> 
> i'm using Rational Apex v3 and have tried a mini version at home (using
> the above) and ObjectAda seems to pass it by reference (not 100% on this
> as my Intel assembler knowledge is zilch!).

ObjectAda would pass such an object by reference unless the actual
were misaligned (maladjusted? ;-), marked atomic or volatile, or had
a C_Pass_By_Copy convention.

-- 
-Tucker Taft   stt@averstar.com   http://www.averstar.com/~stt/
Technical Director, Distributed IT Solutions  (www.averstar.com/tools)
AverStar (formerly Intermetrics, Inc.)   Burlington, MA  USA