Adding UNION/MAP -- Feedback and tips pls!

[N.M. Maclaren]

On Mar 2 2013, Janus Weil wrote:
>>
>> I have to confess that my first reaction was that we shouldn't be
>> doing STRUCTURE/UNION at all.
>
>We certainly shouldn't be doing STRUCTURE. This is really just the
>same as TYPE in standard Fortran.
>
>However, UNION looks like a reasonable extension to me (which has no
>direct replacement in standard Fortran - unless you count
>EQUIVALENCE). I also think we should only support the standard '%' as
>a component selector (also with unions).

There are good reasons for Fortran not standardising unions and going
for CLASS(*) - unchecked unions are incompatible with strong typing.
My vague recollection is that DEC UNION is unchecked, and it certainly
allows the preservation of overlaid data, which is equally horrible.

>> However, I would be content if it lived
>> behind an option.
>
>Under the above circumstances, I don't really see a problem with
>enabling it by default (i.e. with -std=gnu) as we do for other
>extensions.

Unless people start to use it for new code :-(

> I guess one could even argue whether having -std=gnu by default is
> reasonable at all, or whether the default shouldn't rather be
> -std=f2008ts (possibly augmented by -fall-intrinsics) ...

I definitely feel that it should NOT include major gotchas like the
abominable Cray pointers.  When we were collaborating with Hitachi,
they said that they would implement them if I found find a specification.
I eventually said that I couldn't :-(  I gave up after I found 6 wildly
different specifications, with DEC having several - I didn't even try
to check which compilers with apparently the same specification behaved
similarly.  And I have seen a LOT of trouble with that over the decades.

My justification for that rant is that (most) unions are every bit
as bad.  They should NOT be encouraged.


Regards,
Nick Maclaren.