Re: [boost] [function_type] Tag types documentation

From: Tobias Schwinger (tschwinger_at_[hidden])
Date: 2005-07-03 04:51:18

Dave Gomboc wrote:
>>Tag Types:
>>----------
>>
>>The Function Type library uses tag types to represent one 
>>more attributive 
>>properties of a type, such as its variadicness or its calling 
>>convention.
> 
> 
> "one more"?  Did you mean "one or more"?

Yes. It got lost, somehow...

> 
> "a type"?  It might be one type per tag, but tagging can be applied
> to more than one type, so this should be rephrased.

Well, not at once. It's a 1..* relation:

   one type --- N tags to describe its attributes

This all can happen M times, so we have a "infinity != infinity * infinity" 
situation here. This is why I'm not sure if making it plural is a good here.

> 
> "variadicness"?  I think you meant "variadicity".

I'm not sure "variadicity" is a word either. Sorry, I copied this from David 
Abrahams' proposal assuming it was correct.

I'll use "whether it is variadic" instead...

> 
> I'm not thrilled by "attributive properties" either.  My
> discomforture with this phrase is due to its perceived redundancy.
> Is it common for properties to exist that attribute nothing?
> 

I recently excluded the "decoration" property (pointer, reference,member pointer 
or plain function type) from this set (so "attributive" is meant as opposed to 
"fundamental"). I agree it's not perfect.

s/attributive properties/attributation/

> 
> I propose:
> 
> The Function Type library uses tag types to denote properties of
> types.  For example, the tag
> 
>     tag<fastcall, variadic>
> 
> may be used to denote use of the fast calling convention in
> combination with variadicity.
> 

"Denote" (at least the first context you use it in) doesn't quite cut it because 
it isn't limited to notation. The second problem I have with your proposal is, 
that you are choosing a difficult example before showing a simple one. "One or 
more" needs emphasis.

    The Function Type library uses tag types to represent a type's attributation,
    such as its calling convention or wether it is variadic.

    Tags that represent the values of a single property are called property tags.
    These tags can be used to determine whether one property of a type has a
    particular value.

      is_function< T, variadic >
      is_function< T, default_call >

Does it work?

> 
> 
>>When several tags for the same property appear in the 
>>argument list, only the last 
>>one is used; others are ignored.
>>
>>     tag<nonvariadic,variadic> // same as 'variadic'
> 
> 
> The semantics of tag<nonvariadic, variadic> should be identical to
> the semantics of tag<variadic, nonvariadic>.  It strikes me as
> erroneous that this combination of denotations would not yield a
> compilation error.  Have you a well-founded argument for this
> choice?

I have. Try these use cases:

      // a remove constness (of the pointee) from a member function pointer type T
      function_type<signature<T>,tag<signature<T>,non_const> >::type

      // shortcut:
      function_type<T, non_const >::type

or

      // create a (possibly decorated) function type's variadic version
      function_type<signature<T>,tag<signature<T>,variadic> >::type

      // shortcut:
      function_type<T, variadic >::type

Well, we /could/ only allow overriding only in this very context, however, it 
would make things inconsistent and more complicated to explain, IMO.
I don't see a problem in allowing it either, because most error situations are 
pretty obvious

      a = 1;
      a = 2; // <-- should this line fail to compile?

plus 'tag' seldom has to be used directly by the user, anyway.

Further there is the technical side of the story:

      tag<A,B>

only describes a type (talking C++ type system context here) and thus does not 
trigger a template instantiation. So the error message you get would probably 
involve a lot of trace (to where its PoI is) output and be questionable useful, 
because of this.

> 
> I'm new to this thread; apologies if I'm rehashing something that
> was settled earlier.
> 

No need to apologize. Your feedback is very welcome!

Further some reshing is currently a good thing because there have been 
comparatively drastic changes.

Thanks,

Tobias