!standard 8.3(2)                                    15-01-19  AI12-0094-1/03
!standard 8.3(26/3)
!class binding interpretation 13-11-12
!status Corrigendum 1-2012 14-11-12
!status WG9 Approved 15-06-26
!status ARG Approved 7-0-1  14-10-18
!status work item 13-11-12
!status received 13-11-05
!priority Medium
!difficulty Easy
!qualifier Omission
!subject An access_definition should be a declarative region
!summary

An access_definition is a declarative region.

!question

Is the following legal?

type T is
 access function (A : Integer) return
 access function (A : Float) return Boolean;

One might expect that it is legal, but according to RM "8.1 Declarative Region"
there is a declarative region for T and two declarations of A immediately
within it, which is illegal by 8.3(26/2).

Should we change the definition of "declarative region" to include an
access_definition? (Yes.)

!recommendation

(See Summary.)

!wording

Add after 8.1(2):

     * an access_definition;

!discussion

This clearly seems like an oversight.

GNAT allows this example (it thinks it is legal).

!corrigendum 8.1(2)

@dinsa
@xbullet<any declaration, other than that of an enumeration type, that is not
a completion of a previous declaration;>
@dinst
@xbullet<an @fa<access_definition>;>

!ASIS

Not my job.

[Editor's note: :-). I don't know ASIS well enough to know if this has a
visible effect or not, so I would have left it as **TBD and asked J-P.]

!ACATS test

Create an ACATS C-Test like the original question and like the example in
the AARM note 8.1(18.r-t/4), but a rather low priority to do so.

!appendix

From: Bob Duff
Sent: Tuesday, November 5, 2013  7:13 AM

Maxim Reznik (reznik@adacore.com) noticed this issue.

Is the following legal?

type T is
 access function (A : Integer) return
 access function (A : Float) return Boolean;

I'd think it should be legal, but according to RM "8.1 Declarative Region"
there is a declarative region for T and two declarations of A immediately
within it, which is illegal by 8.3(26/2).

Should we change the definition of "declarative region" to include an
access_to_subprogram_definition?

GNAT says it's legal.

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

From: Tucker Taft
Sent: Tuesday, November 5, 2013  8:54 AM

> ... Should we change the definition of "declarative region" to include 
> an access_to_subprogram_definition?

Yes, I would agree with that change.

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

From: Steve Baird
Sent: Tuesday, November 5, 2013  11:27 AM

Sounds right to me too.

Without this change there would also be questions about factored lists, as in

    procedure P (X, Y : access procedure (Z : in out Integer));

Is there one Z here or two? Pick your poison; without the proposed change
either answer has unpleasant consequences (either the normal equivalence breaks
down or we reject a case where the identifier Z only occurs once (textually)
in the source).

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

From: Erhard Ploedereder
Sent: Tuesday, November 5, 2013  4:04 PM

> Should we change the definition of "declarative region" to include an 
> access_to_subprogram_definition?

Yes, of course. Clearly an oversight in the LRM.

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