Version 1.4 of ai05s/ai05-0071-1.txt
!standard 12.5.1(23.1/2) 08-07-07 AI05-0071-1/03
!standard 12.6(10)
!class Amendment 07-10-24
!status ARG Approved 6-0-2 08-06-21
!status work item 07-10-24
!status received 07-10-05
!priority Medium
!difficulty Medium
!subject Class-wide operations for formal subprograms
!summary
(See proposal.)
!problem
There is no predefined equality operator for class-wide types. This causes a problem
when instantiating predefined containers in that it is necessary to create a dispatching
equality operation for this purpose:
function Equal(Left, Right : T'Class) return Boolean;
package T_Vectors is new
Ada.Containers.Indefinite_Vectors(
Positive, T'Class, Equal);
!proposal
There seem to be two ways to fix this:
(1) Define an equivalence for the purposes of formal subprogram default
matching so that the class-wide type can make a similar substitution
that it does inside of the generic; or
(2) Define the idea of "call conformance"; this would include the
matching between formals and actuals in a call.
!wording
Modify 12.5.1(23.1/2):
In the case where a formal type [is tagged with] {has} unknown
discriminants, and the actual type is a class-wide type T'Class:
Modify 12.6(10) as follows:
If a generic unit has a subprogram_default specified by a box, and
the corresponding actual parameter is omitted, then it is equivalent
to an explicit actual parameter that is a usage name identical to the
defining name of the formal. {If a subtype_mark in the profile of the
formal_subprogram_declaration denotes a formal private or formal derived
type, and the actual type for this formal type is a class-wide type
T'Class, then for the purposes of resolving this default_name at the
point of the instantiation, for each primitive subprogram of T that has
a matching defining name, that is directly visible at the point of the
instantiation, and that has at least one controlling formal parameter,
a corresponding subprogram with the same defining name is directly
visible, but with T systematically replaced by T'Class in the types
of its profile. The body of such a subprogram is as defined in
12.5.1 for primitive subprograms of a formal type when the actual
type is class-wide.}
!discussion
In developing a spec for the "tagged" version of the ASIS "Views"
packages, there emerged an "interesting" annoyance with the
Indefinite_Vectors container. Because it takes "=" as a formal
subprogram parameter, if one instantiates it with a class-wide type,
there is no "=" automatically available. It is necessary to explicitly
declare an "Equal" function on the class-wide type, and pass it in. The
body of the "Equal" function would simply make a dispatching call on the
"=" for the correspnding specific type. E.g.:
function Equal(Left, Right : T'Class) return Boolean;
package T_Vectors is new
Ada.Containers.Indefinite_Vectors(
Positive, T'Class, Equal);
It is important not to actually name it "=" because
if so named, it would create ambiguity at every call.
This all seems kind of annoying/surprising. Since one is allowed in an
instantiation to provide a class-wide type if the formal type has
unknown discriminants, and there are new rules for how the primitives of
the formal type are to be defined in terms of the actuals (paras
23.1/2-23.3/2 of 12.5.1), it would seem to make sense for such
operations to be available for formal subprograms with a "<>" default.
Hence, we have proposed wording in 12.6 to allow for these operations,
and to refer to 12.5.1 for the definition of the semantics of the body.
The change to the wording in 12.5.1(23.1/2) is to eliminate the word
"tagged," because the formal need not be tagged for the actual to be
class-wide, in the case of a formal private type with unknown
discriminants.
The addition to the wording of 12.6 should allow instantiations with
class-wide types to work as the user would expect. That is, by adding
in to a generic formal part:
with function "="(Left, Right : T)
return Boolean is <>;
we are simply trying to import any user-defined equality operator rather
than having some underlying equality operator reemerge, while also
making it clear that equality is important to the functioning of the
generic. Unfortunately, as it is now, adding in this nice piece of
functionality penalizes instantiations for which the actual is a
class-wide type.
An alternative approach would be to change the rules for matching
"expected profiles," which might imply having a new kind of conformance
between profiles, such as "call conformance." That is "call
conformance" would include the matching allowed between formal parameter
types and actual parameter types in a call (8.6(20/2-25.1/2)). This
would obviously be a bigger change, and we have chosen to go with the
smaller change in this AI.
!example
generic
type Priv(<>) is private;
with function "="(Left, Right : Priv)
return Boolean is <>;
X, Y : Priv;
package GP is
Are_Equal : constant Boolean := (X = Y); --
end GP;
...
procedure Test(U, V : T'Class) is
use type T;
package P is new GP(T'Class, U, V);
--
--
--
begin
if P.Are_Equal then
...
end if;
end Test;
!corrigendum 12.5.1(23.1/2)
Replace the paragraph:
In the case where a formal type is tagged with unknown discriminants,
and the actual type is a class-wide type T'Class:
by:
In the case where a formal type has unknown discriminants,
and the actual type is a class-wide type T'Class:
!corrigendum 12.6(10)
Replace the paragraph:
If a generic unit has a subprogram_default specified by a box, and the
corresponding actual parameter is omitted, then it is equivalent to an explicit
actual parameter that is a usage name identical to the defining name of the formal.
by:
If a generic unit has a subprogram_default specified by a box, and the
corresponding actual parameter is omitted, then it is equivalent to an explicit
actual parameter that is a usage name identical to the defining name of the formal.
If a subtype_mark in the profile of the
formal_subprogram_declaration denotes a formal private or formal derived
type, and the actual type for this formal type is a class-wide type
T'Class, then for the purposes of resolving this default_name at the
point of the instantiation, for each primitive subprogram of T that has
a matching defining name, that is directly visible at the point of the
instantiation, and that has at least one controlling formal parameter,
a corresponding subprogram with the same defining name is directly
visible, but with T systematically replaced by T'Class in the types
of its profile. The body of such a subprogram is as defined in
12.5.1 for primitive subprograms of a formal type when the actual
type is class-wide.
!ACATS test
ACATS C-Test(s) are necessary for this change.
!appendix
From: Tucker Taft
Date: Friday, October 5, 2007 11:21 AM
In developing a spec for the "tagged" version of the
ASIS "Views" packages, I ran into an "interesting"
annoyance with the Indefinite_Vectors container. Because
it takes "=" as a formal subprogram parameter, if
one instantiates it with a class-wide type, there is
no "=" automatically available. I had to explicitly
declare an "Equal" function on the class-wide type,
and pass it in. The body of the "Equal" function would
simply make a dispatching call on the "=" for the
correspnding specific type. E.g.:
function Equal(Left, Right : T'Class) return Boolean;
package T_Vectors is new
Ada.Containers.Indefinite_Vectors(
Positive, T'Class, Equal);
I also had to be sure I didn't actually name it "=" because
if I did so, I would create ambiguity at every call.
That seems kind of annoying/surprising. Since we
now acknowledge the possibility of instantiating a
formal type with unknown discriminants with a class-wide
type, and have defined the rules for how the
primitives of the formal type are to be defined
in terms of the actuals (paras 23.1/2-23.3/2 of 12.5.1),
it would seem to make sense to also make such
operations available for formal subprograms
with a "<>" default. This would imply adding another
paragraph after 23.3/2, something like:
* If a primitive operation of T is directly visible
at the point of the instantiation, then a subprogram
with the same name, but with T systematically replaced
by T'Class in formal parameter types and result types,
is considered directly visible at the point of the
instantiation, for the purpose of resolving the
actual parameter for a formal subprogram with a
subprogram_default specified by a box. The body for
such an operation, and controlling tag determination
for such an operation, is as defined above.
Although this a bit of a mouthful, I believe it is reasonable
if we are trying to allow instantiations with class-wide
types to work as the user would expect. That is, by
adding in to a generic formal part:
with function "="(Left, Right : T)
return Boolean is <>;
we are simply trying to import any user-defined equality
operator rather than having some underlying equality
operator reemerge, while also making it clear that equality
is important to the functioning of the generic. Unfortunately,
as it is now, adding in this nice piece of functionality
penalizes instantiations for which the actual is a class-wide type.
An alternative approach would be to change the rules for matching
"expected profiles," which might imply having a new kind of
conformance between profiles, such as "call conformance." That
is "call conformance" would include the matching allowed between
formal parameter types and actual parameter types in a call
(8.6(20/2-25.1/2)). This would obviously be a bigger change.
****************************************************************
Questions? Ask the ACAA Technical Agent