Re: C/C++ programmer giving Ada95 a chance -- writing an emulator.

[Tucker Taft <stt@averstar.com>]

Geoff Bull wrote:
> 
> If I Initialise the array like so:
> 
>    Op_Table : array (Opcode) of Op_Root
>      := (0 => Op_Root (O0),
>          others => OU);
> 
> then make the assignments:
>     Op_Table (1) := Op_Root (O1);
>     Op_Table (2) := Op_Root (O2);
>     Op_Table (3) := Op_Root (O3);
> 
> I get the output:
> 
> opcode 00
> undefined opcode
> undefined opcode
> undefined opcode
> undefined opcode
> 
> What is going on here?
> It look like the assigments throw away the tag (like Cohen
> says it should) but the initializer keeps the tag.

That would be a bug.  The tag is *not* determined by the initial
value if the target type is "specific" (as opposed to classwide).
The tag *is* determined by the initial value if the target type is classwide.
 
> Indeed the RM says in section 3.9 (22):
> "The tag of an object of a class-wide tagged type is that of its
> initialization expression. "
> The trouble is Op_Table components are not of a class wide type.
> 
> Instead section 3.9 (20) applies:
> "The tag of a stand-alone object, a component, or an aggregate of a
> specific tagged type T identifies T"
> 
> >   I think we have here both an error in Gnat and an error in
> > Cohen's book!
> 
> Unless somebody can show me where the RM says otherwise, I agree.
> I guess I'd better report this.

Definitely a bug.  It looks like GNAT is copying the tag from the initial value,
which it shouldn't do in this case.
> 
> Thanks for the enlightenment.
> Cheers
> Geoff

-Tucker Taft  stt@averstar.com