Re: [boost] [contract] toward N3351 concepts

Subject: Re: [boost] [contract] toward N3351 concepts
From: Andrew Sutton (asutton.list_at_[hidden])
Date: 2012-09-27 10:29:57

I'm glad to see this discussion.

>> > I believe that N3351 is not a good place to learn axioms from.

It's a reasonable place to start as long as you don't expect too much
from them. The axioms presented in that paper were primarily intended
to document the semantics of a type; we had not planned (at that time)
any particular ideas about how they should or could be used by the
compiler.

> (2). Axioms need not hold for all values of the given type, but only for
> those values that a given algorithm (constrained by a concept with axioms),
> and only in expressions that are really stated in the executed algorithm.

Axioms, in the context of a concept, are intended to hold for all
values of type. A precondition is a property of specific values or
objects and is not an axiom.

> For instance, axiom EqualityComparable::Reflexivity states that some
> (rather than every) objects must compare equal to themselves. If you
> consider type float, objects with value NaN (not-a-number) compare unequal.

No; all well-formed values of a type can be compared for equality (if
the type is EqualityComparable). NaN is not a well-formed value. It is
literally *not a number*.

Andrew