Re: 3-bit array ...

[Tucker Taft <stt@averstar.com>]

Robert Dewar wrote:
> 
> In article <wccg0xgm5n4.fsf@world.std.com>,
>   Robert A Duff <bobduff@world.std.com> wrote:
> > GNAT has always
> > supported packing to the 3-bit level.  AverStar's front end,
> and the
> > Green Hills compiler that uses AverStar's FE, used to use 4,
> but now
> > uses 3, due to customer demand for it.  That's what this whole
> > discussion has been about.
> >
> > There's also the issue of whether you can say:
> >
> >     for T2'Component_Size use 3;
> >
> > instead of the pragma Pack above.
> 
> I must say I wonder if the GNAT (and now Averstar) decision is
> the right one. It would have been cleaner to say Pack always
> means 4, and you use a component size clause if you want 3.

Bob mis-spoke.  We only pack to "odd-ball" sizes if the user
explicitly requests it via a 'Component_Size clause.  This
is important because there is existing Ada 95 code which we didn't
want to suddenly change behavior in a potentially surprising way.

> But the gods of legacy Ada 83 code definitely demand significant
> sacrifices, so this is one of them.

We chose to force these "legacy" users to put in a 'Component_Size clause
if they really need the 3-bit (in)efficiency.

> Bob, how far did Averstar go, did you just do 3 bits? In GNAt
> we close pack all sizes up to 64. (e.g. an array of Natural
> gets packed to 31, which can be quite a surprise, although the
> pragma Pack in this case is junk!)

We support any "odd-ball" 'component_size up to the maximum integer 
size supported by the compiler on the target.  That varies, from 24 
bits up to 64 bits.

-- 
-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