To: J3                                                     J3/22-172
From: Malcolm Cohen
Subject: Conditional argument description
Date: 2022-July-18
Reference: 22-155


0. Introduction

This paper is the response from DATA subgroup to 22-155.


1. Properties

22-155 states
  "a <conditional-arg> is not an object so we feel that it is not
   appropriate to give it properties that objects have including type, kind
   type parameters, and rank"

This is fundamentally misguided. X+1 is not a data object either,
nonetheless it has type, type parameters, and rank.

Similarly, components of a type are not data objects either, or even data
entities. ("structure components" are data objects, but not components of
a type.)

In fact "an expression" is not even a data entity in itself, even though
the result of evaluating it is one. That does not stop an expression from
having type, type parameters, and rank, in fact we have whole subclauses
about what those are.

Therefore this objection seems groundless.
(Further, it would be peculiar for conditional expressions to have type et
al, but conditional arguments not, even though they are almost the same
syntax and were deliberately designed in parallel.)


2. Generic resolution

22-155 claims
    "Note that this does not remove the property of generic resolution
     being independent of the choice of a <consequent-arg>."

However, if a conditional argument has no type, it cannot be used for
generic resolution. Probably a minor rewording would be needed here if
we followed this change.


3. Alternative syntax

This appears to have no advantages over the current syntax, and appears
to be more complicated. It is unclear what this is intended to achieve.


4. Conclusion

No errors in the existing text have been identified, and this alternative
approach is not obviously superior.

Thus no action on 22-155 is recommended.

===END===