Re: [boost] Formal Review: Concept Checking Library

From: Beman Dawes (beman_at_[hidden])
Date: 2000-11-24 20:20:02

At 01:03 PM 11/21/2000 -0500, Jeremy Siek wrote:

 >On Mon, 20 Nov 2000, Matthew Austern wrote:
 >>  - I keep hoping that there might be a way to eliminate the
 >>    use of macros.  Token pasting, in particular, is distasteful.
 >
 >Here's my first stab at a non-macro approach:
 >
 >First, for the function requires we can just use a function
 >template like this:
 >
 >template <class Concept>
 >void function_requires()
 >{
 >  void (Concept::*x)() = BOOST_FPTR Concept::constraints;
 >  ignore_unused_variable_warning(x);
 >}
 >
 >And then use explicit template arguments to invoke it:
 >
 >boost::function_requires< DefaultConstructibleConcept<foo> >();

That's fine.

 >Second, for class requires we can create a class template
 >like this:
 >
 >template <class Concept>
 >struct class_requires
 >{
 >  typedef void (Concept::* function_pointer)();
 >
 >  template <function_pointer Fptr>
 >  struct dummy_struct { };
 >
 >  typedef dummy_struct< BOOST_FPTR Concept::constraints > check;
 >};
 >
 >and then invoke it like this:
 >
 >class class_requires_test
 >{
 >  typedef class_requires< EqualityComparableConcept<int> >::check req1;

That's not so good.  The "typedef", "::check", and "req1" detract from the 
clarity of the code.  I worry developers will shy away from using concept 
checks because of the ugly code.

Another worry: in testing an early compile time asserts proposal which used 
a similar typedef technique, I found a number of slight coding mistakes 
would compile OK, but not actually do the indicated check.  What happens if 
a user codes any of these mistakes (where X is a UDT that isn't Equality 
Comparable)?

     class_requires< EqualityComparableConcept<X> >::check req1;
     class_requires< EqualityComparableConcept<X> >::check;
     class_requires< EqualityComparableConcept<X> > req2;
     class_requires< EqualityComparableConcept<X> >;
     typedef class_requires< EqualityComparableConcept<X> > req3;
     typedef class_requires< EqualityComparableConcept<X> >;

If they fail to compiler, fine, but if they compile without error, and fail 
to detect the error, I would be concerned.  If they did error check OK, but 
generated code because the typedef was missing, that would also be a 
concern. Could you give them a try?

While the macros were, well, macros, they were also very clear, and seemed 
hard to misuse IMO.

--Beman