Re: calenday (was Re: IBM Acquires Rational Ada

["Marin David Condic" <mcondic.auntie.spam@acm.org>]

Not at all true. Many message protocols look very much like an Object
Oriented thing with a base class and many derived classes. It is extremely
natural to represent this as a tagged record. Not so much because you expect
to transmit the tag itself, but because you expect to process your messages
internally with all sorts of dispatching and inheretance and stuff like
that. It is extremely natural to do this in an OO way. What you need to do
though is to control the representation of everything that comes after the
tag. You look at your communication link and once you determine the
appropriate type of message, you're doing an overlay or unchecked conversion
or something similar to get the bytes into the tagged record and then go do
your dispatching from there.

The thing is, without the tagged record, you end up building your own
dispatching and inheritance and things like that anyway. (Big case
statements, similarly named procedures or procedures common to all message
types, etc.) You're either going to do it with a lot of bit-twiddling and
extra work or you can do it with tagged records very cleanly. If you've got
two Ada programs talking to each other and they're built by the same
compiler, representation can become a "don't care" issue - maybe. If its on
a more heterogeneous plane, you've got to be sure you control
representation. So I don't really think the issue is all that abstract or
theoretical - it does occur in the real world often enough and you'd just
like to take advantage of tagged records & representation to make it work
nicely.

MDC
--
======================================================================
Marin David Condic
I work for: http://www.belcan.com/
My project is: http://www.jast.mil/

Send Replies To: m c o n d i c @ a c m . o r g

    "I'd trade it all for just a little more"
        --  Charles Montgomery Burns, [4F10]
======================================================================

Frank J. Lhota <NOSPAM.lhota.adarose@verizon.net> wrote in message
news:zuKM9.66886$4W1.13625@nwrddc02.gnilink.net...
> "Hyman Rosen" <hyrosen@mail.com> wrote in message
> news:1040410199.497867@master.nyc.kbcfp.com...
> > The whole concept of representation specs on tagged
> > records seems pretty dodgy to me.
>
> I agree. AFAIK, the only reason to use a representation clause is to make
> sure that a type matches some external data layout. Tagged types, by their
> very nature, do not fit into this category. This seems to be one of those
> theoretical issues with no practical consequences.
>
>