Re: Aliased Unchecked_Unions not seen or taken notice of

["Randy Brukardt" <randy@rrsoftware.com>]

"Lucretia" <laguest9000@googlemail.com> wrote in message 
news:bed99006-94ac-457c-9461-24d326eb4c60@googlegroups.com...
> On Monday, 12 November 2018 20:13:53 UTC, Dmitry A. Kazakov  wrote:
>> On 2018-11-12 19:29, Lucretia wrote:
>> > On Monday, 12 November 2018 17:56:05 UTC, Lucretia  wrote:
>> >
>> >>     type Results (Success : Boolean) is
>> >>     record
>> >>        case Success is
>> >>           when False =>
>> >>              Error : Errors_Ptr;
>> >>
>> >>           when True =>
>> >>              Output : Values_Ptr;
>> >>        end case;
>> >>     end record with
>> >>       Convention      => C_Pass_By_Copy;
>> >
>> > This compiles fine. So it seems impossible to use unchecked_union and 
>> > access types together.
>>
>> What about C_Pass_By_Copy? That looks much inconsistent with access to 
>> me.
>
> I did try removing that too.
>
>> BTW, why do you want to use unchecked union? In comparable cases I
>> rather simply overload imported functions with whatever arguments:
>
> Because it's not that simple.

Sure, but it's almost always simple enough. We built Claw without using 
Unchecked_Union at all (it didn't exist in Ada 95), and there weren't many 
(if any) cases where the result would have been better had it existed. 
Worse, Unchecked_Union is an easy way to introduce erroneous execution into 
one's program accidentally; at least using Unchecked_Conversion or some 
similar scheme makes that obvious. (Using the "wrong" discriminant is 
erroneous, even though that isn't checked or checkable.)

I'd keep the use of Unchecked_Union *very* limited, mostly to just cases 
where it is a component of some outer type. And even that might be better 
handled with Unchecked_Conversion (for reasons noted above).

                                               Randy.