Re: [boost] any_function: bcc 5.5.1; explicit ctor

From: Doug Gregor (gregod_at_[hidden])
Date: 2001-03-28 12:03:52

On Wednesday 28 March 2001 02:59, you wrote:
> 1. I have just downloaded the most recent version of any_function, which
> documentation says it compiles under bcc 5.5.1, but it doesn't! The
> following simple code fails to compile:
>
> 	#include <boost/any_function.hpp>
>
> 	void foo(int) {}
> 	int main()
> 	{
> 		boost::any_function<void, int> v;
> 		v = foo;
> 	}
>
> The problem seems to be in using enums for constants in various places.
> Consider:
>
> 	template<int i> struct I { enum { val = i }; };
> 	template<int i> struct V {};
>
> 	template<int i>
> 	struct S {
>
> 		enum { val = I<i>::val };
>
> 		typedef V<val> type;     //  1
> 	};
>
> Given this declarations, bcc thinks that S<23>::type is ... V<0>! This is
> cleary bug, which affects any_function.hpp. I'm not sure if it can be
> workarounded, leaving enums, but one can change val to static const int
> member, and change line #1 to
> 	typedef V<int(S::val)> type;
> This will work. I actully wonder why enum is used for const values. Is
> there any other compiler that has problems with static const ints?

Sorry, foolish oversight on my part. BOOST_STATIC_CONSTANT is meant to deal 
with this (MSVC is one of the compilers that can handle enums but not static 
const ints). 

> 2. I found explicit ctor of any_function not very good idea.
> 	-	It adds no safety at all, but is a big encumbrance.
> 	-	Is it reasonable to make ctor explicit while permitting assigment
> 		without explicit conversion? Don't you think that if conversion of type
> 		A to type B is considered dangerous, then both marked lines should not
> 		compile?
> 			struct S {
> 				S();
> 				S(const B& b) : b(b) {};
> 				B b;
> 			};
> 			A a;
> 			S s1(a);		//*
> 			S s2;
> 			s2.b = a;	//*

Thank you for noticing this, I'd forgotten. In a prior incarnation of 
any_function, I had been running into compiler problems with non-explicit 
constructors of the form:

template<typename T> class_name(T);

However, this does not appear to be the case currently, and it makes no sense 
to disallow implicit conversions to any_function. 

I've uploaded an updated version of any_function with the 
BOOST_STATIC_CONSTANT fix and with implicit conversions from any function 
object. 
Simple test programs run properly on VC++ 6, BC++ 5.5.1, GCC-2.95.2, GCC-3.1 
(prerelease), and Comeau 4.2.44b3.

The newest version is available at:
http://groups.yahoo.com/group/boost/files/Callback/any_func-03-28.zip

Finally, I'll be leaving town Friday and won't be back for a week, so I 
apologize in advance if bug reports for any_function go untouched.

        Doug