!standard 10.1.1(12.3/2)                                          08-07-12  AI05-0108-1/01
!class binding interpretation 08-07-12
!status work item 08-07-12
!status received 06-05-30
!priority Low
!difficulty Easy
!qualifier Omission
!subject The incomplete view from a limited view does not have a discriminant part
!summary

The incomplete view of a type found in a limited view of a package does not have
a discriminant part even if the full view of the type does have a discriminant
part.

!question

10.1.1(12.3) discusses what the limited view of a package looks like, with
respect to types declared in that package:

    For each type_declaration in the visible part, an incomplete view
    of the type; if the type_declaration is tagged, then the view is a
    tagged incomplete view.

Does this incomplete view include a discriminant part, if the type has one? (No.)

Since incomplete views (declared by incomplete_type_declarations) may have
discriminant parts, the above language doesn't seem to exclude the
possibility of a discriminant part, but it also doesn't imply that a
discriminant part is present.

So, is the following legal? (No.)

    package Pak1 is
       type T1 (Disc : Integer) is record ... end record;
    end Pak1;

    limited with Pak1;
    package Pak2 is
       type Acc_T1 is access Pak1.T1(4); -- Error.
    end Pak2;

!recommendation

(See summary.)

!wording

Change 10.1.1(12.3/2):

For each type_declaration in the visible part, an incomplete
view of the type {with no discriminant_part}; if the type_declaration
is tagged, then the view is a tagged incomplete view.

AARM Ramification: The incomplete view of a type does not have
a discriminant_part even if the the type_declaration does have
one. This necessary because semantic analysis (and the associated
dependence on with_clauses) would be necessary
to determine the types of the discriminants.

!discussion

One could argue that since the wording says nothing about discriminant_parts,
one is not present. (Indeed, the author initially did just that.) However,
one could easily imagine that such a part is present; it's just not clear.

However, it should be clear from the model of limited views that such 
discriminant_parts cannot be present. A limited view is intended to be
creatable only with syntactic analysis: in particular, no name resolution
is required. 

The example in the question clearly violates that: without semantic
analysis, the meaning of the name "Integer" is not known. Moreover, the
model is that all types are incomplete. That would prevent giving useful
discriminant constraints anyway: the value "4" is not a value of any
incomplete view.

Thus the best thing to do is to add wording to clarify the intent.

!corrigendum 10.1.2(12.3/2)

@drepl
@xbullet<For each @fa<type_declaration> in the visible part, an incomplete
view of the type; if the @fa<type_declaration> is tagged, then the view
is a tagged incomplete view.>
@dby
@xbullet<For each @fa<type_declaration> in the visible part, an incomplete
view of the type with no @fa<discriminant_part>; if the @fa<type_declaration>
is tagged, then the view is a tagged incomplete view.>

!ACATS Test

An ACATS B-Test should be created to check that this is illegal. (Probably
just add cases to the existing tests.)

!appendix

!topic RM doesn't say whether incomplete views have discriminants
!reference 10.1.1(12.3)
!from Adam Beneschan 08-05-30
!discussion


10.1.1(12.3) discusses what the limited view of a package looks like, with
respect to types declared in that package:

    For each type_declaration in the visible part, an incomplete view
    of the type; if the type_declaration is tagged, then the view is a
    tagged incomplete view.

However, nothing says whether the incomplete view includes a discriminant part
or not, if the type has one.  Since incomplete views (declared by
incomplete_type_declarations) may have discriminant parts, I cannot see how
the above language excludes the possibility of a discriminant part, but it
also doesn't imply that a discriminant part is present.  So this seems
like a serious omission in the RM.  In particular, I can't tell whether this
is legal:

    package Pak1 is
       type T1 (Disc : Integer) is record ... end record;
    end Pak1;

    limited with Pak1;
    package Pak2 is
       type Acc_T1 is access Pak1.T1(4);
    end Pak2;

Skimming through AI95-217-6, it appears that the intent was that discriminant
parts were not intended to be included on types in limited views of packages
(unless this got changed later by AI-326, or unless I missed something in
AI95-217).  But it appears that the language that would say so didn't make it
into the RM.

I think 10.1.1(12.3) needs to say something one way or the other.

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