!standard  3.5                                      06-11-06    AC95-00136/01
!class Amendment 06-11-06
!status received no action 06-11-06
!status received 06-07-19
!subject Ada should have elementary sets

!summary

!appendix

[Editor's note: This appeared in an ASIS discussion, but has been
filed separately because it has nothing at all to do with ASIS.]

From: Tucker Taft
Date: Wednesday, July 19, 2006  6:46 PM

CONTINUING OFF TOPIC...

I'll admit I wish I could often use "choice_list" syntax
in membership tests, e.g.:

     if G(X) in A..B | D..F then ...

and using such notation in subtype definitions
would also seem to make sense.

Robert Dewar wrote:
> Tucker Taft wrote:
>> I have certainly used ranges with enumeration types,
>> but that only works if the enumeration type is
>> ordered in a way that corresponds to a depth-first
>> walk of some kind of hierarchical structure.
>
> and then works only partially, I must say it is a big
> weakness of Ada that subtypes of enumerations have to
> be contiguous :-(

****************************************************************

From: Robert Dewar
Date: Wednesday, July 19, 2006  10:02 PM

I definitely agree, perhaps we will add this as an extension :-)

****************************************************************

From: John Barnes
Date: Thursday, July 20, 2006  1:32 AM

I agree too and have done so since 1982 when first writing my book. Exercise
7.2(2) has an enumeration type

type Month_Name is (Jan, Feb, Mar, ..., Dec);

and then I want subtypes for the seasons. I actually write

subtype Winter is Month_Name range Jan .. Mar; and so on.

But that feels wrong. What I really want is winter to be December, January
and February and so on.

Oh dear.

****************************************************************

From: Robert I. Eachus
Date: Saturday, July 22, 2006  9:53 PM

I'm going to say this, smile, then run and hide.  If someone wants to
propose this functionality for a future version of Ada, I suggest we add
sets.  Make *set* a reserved word, and allow users to declare sets like
subtypes and use them in most of the same places.  That would flag to
the user that he was possibly making lots of work not for the compiler,
but for the computer at run-time.  It is probably worth going a bit
further, and allowing values to be added to or removed from sets.  You
could even declare constant sets that could not be modified:

constant set Digits is Character range '0'..'9' and 'A' ..'F'; --or some
such.

I know how easy it is to use Boolean types to implement sets in Ada, and
I have done it lots of times.  But it may be time to grasp the nettle,
and add syntactic sugar that the compiler can convert.

Of course someone is sure to point out that with all this, why not allow
sets over access, tagged, and record types?

As I said running and hiding. ;-)

****************************************************************

From: Stephen Michell
Date: Monday, July 24, 2006  3:03 AM

I was going to write almost exactly that note last week, but I ran and
hid before I sent it.

I fully support adding sets, and not doing the range splitting that has
been suggested. Dscontiguous ranges are only one of the applications of
sets.

The most significant issue, in my opinion, is the name choice. "set" is
the most obvious choice, and probably the only choice, but is likely a
type name in 1/2 the Ada programs in the worls.

****************************************************************

From: Robert Dewar
Date: Wednesday, July 26, 2006  1:31 AM

To me the range splitting is a simple extension, syntactically and
semantically, and we even think of doing this as an extension. But
adding sets is a very big language design step, I doubt it will ever
happen!

One related issue we have considered is a pragma that would say that
an enumeration type is unordered, and prohibit any constructs that
depend on the ordering (incoluding comparison and subranges).

****************************************************************