Re: [boost] [type_erasure] Some issues

Subject: Re: [boost] [type_erasure] Some issues
From: Steven Watanabe (watanabesj_at_[hidden])
Date: 2013-02-25 18:40:24

AMDG

On 02/25/2013 11:30 AM, Fabio Fracassi wrote:
> On 2/25/13 6:13 PM, Steven Watanabe wrote:
> [snip]
>>> First I think I discovered a bug either in your lib or in clang(with
>>> libc++) when compiling one of the documentation examples in C++11 mode.
>>>
>>
>> I haven't tested much with clang in C++11 mode.
>> I wasn't able to get libc++ working on Linux.
>> Are you using the latest version: r83101?
>> Also, exactly what version of clang are you
>> using?  I'll take a look at this later today.
>>
> 
> Just updated to revision 83153 (boost sandbox), The problem remains.
> I am using
> Apple LLVM version 4.2 (clang-425.0.24) (based on LLVM 3.2svn)
> <snip>
> 

I've just committed a fix for this.  I just
forgot to handle a case inside one clang
#ifdef branch.

> 
> [snip]
>>
>> The problem is the way name lookup works.  It's clear that
>> draw can't be found in the template definition, since it
>> isn't declared until later on.  Argument dependent lookup
>> usually works, but in this case, the only associated namespace
>> is std (from std::ostream) and draw is defined in the global
>> namespace. (Note that the built-in types have no associated
>> namespaces).  BOOST_TYPE_ERASURE_FREE will have exactly
>> the same problem.  The solution is add declarations of draw
>> for all builtin types /before/ the definition of drawable.
> 
> Ok, this works, but makes it impossible to (reliably) define the concept
> in isolation.
> Quite a severe limitation. Is there any possibility to overcome this?
> 

This isn't as bad as you might think.
It only applies to built in arithmetic
types and types in the standard library
of which there are a known fixed set.

In Christ,
Steven Watanabe