Version 1.3 of ai12s/ai12-0162-1.txt

Unformatted version of ai12s/ai12-0162-1.txt version 1.3
Other versions for file ai12s/ai12-0162-1.txt

!standard B.3.3(25/2)          15-07-30 AI05-0162-1/03
!class binding interpretation 15-06-04
!status ARG Approved 8-0-2 15-06-26
!status work item 15-06-04
!status received 15-04-28
!priority Low
!difficulty Easy
!qualifier Omission
!subject Memberships and Unchecked_Unions
!summary
Program_Error is raised if any individual membership test fails the rule of B.3.3(25/2).
!question
B.3.3(25/2) says: "Evaluation of a membership test if the subtype_mark denotes a constrained unchecked union subtype and the expression lacks inferable discriminants."
But Ada 2012 changed memberships to test a membership choice list. There is no "THE" subtype_mark, there could be several. And of course there are potentially several expressions involved as well; we added a prefix "tested_" to clarify that in the membership wording (and it has become a simple_expression to boot).
Should this be reworded? (Yes.)
!recommendation
(See Summary.)
!wording
Modify B.3.3(25/2):
Evaluation of [a] {an individual} membership test if the subtype_mark {(if any)} denotes a constrained unchecked union subtype and the {tested_simple_expression}[expression] lacks inferable discriminants.
!discussion
This wording will cause the membership
Non_Inferable_Expr in Con_UU | Uncon_UU
to raise Program_Error rather than evaluating to True. We could have required that all of the subtypes fail this check before we raise Program_Error, but this would mean that we'd be allowing an individual membership test whose result is undefined, and allowing that because these results are ORed (anything ORed with True is True). That seems unusual and unnecessary as such an expression makes little sense to begin with (why write a membership which is always True? And if so, why combine it with one that requires an actual test?). It's important that these things are well-defined, but how they're defined isn't that important.
We could have talked about the "subtype_mark of a membership_choice" instead of talking about invididual membership tests, but that's just longer, not clearer.
Note that the choice_expression case is covered by B.3.3(23/2) if necessary, as evaluating that choice is defined to be an evaluation of primitive equality for a record type; if there is no user-defined equality, then B.3.3(23/2) applies. (If there is a user-defined equality, then B.3.3(23/2) does not apply and the call is made; the implementation of the user-defined equality might trigger one of these rules itself, but it doesn't have to.)
!corrigendum B.3.3(25/2)
!ASIS
No ASIS effect.
!ACATS test
An ACATS C-Test could check that this rule is properly implemented, but as interesting cases are rather pathological, and simple cases are already tested by test CXB3021, additional tests are very low priority.
!appendix

From: Randy Brukardt
Sent: Monday, April 20, 2015  11:32 PM

I'm working though the objectives for Unchecked_Unions so that I can issue a
bunch of tests submitted by AdaCore years ago. Of course, this means reading
the rules carefully for the first time in forever.

B.3.3(25/2) says:
"Evaluation of a membership test if the subtype_mark denotes a constrained
unchecked union subtype and the expression lacks inferable discriminants."

But Ada 2012 changed memberships to test a membership choice list. There is no
"THE" subtype_mark, there could be several. And of course there are potentially
several expressions involved as well; we added a prefix "tested_" to clarify
that in the membership wording (and it has become a simple_expression to boot).

This should say something like
"Evaluation of a membership test if a membership_choice subtype_mark denotes a
constrained unchecked union subtype and the tested_simple_expression lacks
inferable discriminants."

P.S. I'd love to find a volunteer to write a usage-oriented C-Test for
Unchecked_Unions. (Hint, Hint.) That needs to pass an object of such a type to
some C-code (probably emulating some C API) and receive some answer back that
depends on the union. I think I'd spend longer trying to write the C code for
that than would be healthy. :-) (I can do the Ada part if necessary, it's the
C part I need help with.) The submitted tests mostly try corner cases and
Legality Rules; definitely needed but the primary use should be tested, too.

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

From: Brad Moore
Sent: Tuesday, April 21, 2015  7:21 AM

Im willing to write the c test. It might be a few weeks before I'd get around
to it though.

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

From: Tucker Taft
Sent: Tuesday, April 21, 2015  8:44 AM

...
> B.3.3(25/2) says:
> "Evaluation of a membership test if the subtype_mark denotes a 
> constrained unchecked union subtype and the expression lacks inferable discriminants."
>
> But Ada 2012 changed memberships to test a membership choice list. 
> There is no "THE" subtype_mark, there could be several. And of course 
> there are potentially several expressions involved as well; we added a 
> prefix "tested_" to clarify that in the membership wording (and it has 
> become a simple_expression to boot).

Good catch.

> This should say something like
> "Evaluation of a membership test if a membership_choice subtype_mark 
> denotes a constrained unchecked union subtype and the 
> tested_simple_expression lacks inferable discriminants."

We could say that it returns True if any of the subtype_marks are
unconstrained, but I suppose it is a bit weird to have undefined values
connected with an "or else" with True.  If we decide to return True if any
of the choices are unconstrained, then we should say "every membership_choice
subtype_mark" instead of "a membership_choice subtype_mark".  I could go either
way on this...

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

From: Randy Brukardt
Sent: Tuesday, April 21, 2015  2:52 PM

I thought about that, which was the main reason that this has to be an AI for 
discussion. We need to decide what happens with:

    Expr in Con_UU | Uncon_UU

but that case is pretty pathological (the membership is always True, so why
write it at all, and if you do, why include the constrained one?), so I don't
know if it really matters what we decide (just that we decide something). The
main need for rewording is so that

    Expr in Con1_UU | Con2_UU

is well-defined. But even there, it seems likely that most would do the right
thing.

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

From: Tucker Taft
Sent: Tuesday, April 21, 2015  3:03 PM

> ... The main need for rewording is so that
>
>      Expr in Con1_UU | Con2_UU
>
> is well-defined. But even there, it seems likely that most would do 
> the right thing.

I agree with you, this is all pathological, so your original suggestion seems
simpler, and avoids having to talk about "undefined or else True."

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

From: Jean-Pierre Rosen
Sent: Friday, April 24, 2015  7:00 AM

> This should say something like
> "Evaluation of a membership test if a membership_choice subtype_mark 
> denotes a constrained unchecked union subtype and the 
> tested_simple_expression lacks inferable discriminants."

Do we really need a change? B.3.3(25/2) defines an individual membership test 
(admitedly, the word "individual" is in extremly fine print), and then 27.1/3
applies:

For the evaluation of a membership test using in whose membership_choice_list
has more than one membership_choice, the simple_expression of the membership test
is evaluated first and the result of the operation is equivalent to that of a
sequence consisting of an individual membership test on each membership_choice
combined with the short-circuit control form or else.

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

Questions? Ask the ACAA Technical Agent