Re: [boost] [math_toolkit] ADL question

From: Michael Marcin (mmarcin_at_[hidden])
Date: 2007-06-29 15:39:23

John Maddock wrote:
> Michael Marcin wrote:
>> Hello,
>>
>> I haven't looked much at the math toolkit yet but I was interested in
>> how it handled errors so I peeked at common_error_handling.hpp
>>
>> I see it explicitly qualifies each all to isfinite with
>> (boost::math::isfinite)( value ).
>>
>> I assume this is to prevent macro expansion however this means that
>> if a
>> UDT is passed in the user is required to add their isfinite to the
>> boost::math namespace instead of having it be reachable through ADL.
>>
>> Is this intentional? I think the following would work if it's not.
>>
>> using boost::math::isfinite;
>> isfinite BOOST_PREVENT_MACRO_SUBSTITUTION( value );
>>
>> I'm not sure if UDT support for this library has been considered in
>> its design but I'm working to provide a fixed-point math library that
>> I hope
>> can be used as a real type just like float and double.  I would like
>> to
>> be able to keep as much as possible in the library's namespace.
> 
> Good points, but if we rely on ADL don't we get ambiguities between 
> boost::math::isfinite and ::isfinite (when defined as a global function)?
> 

IIRC the using declaration brings boost::math::isfinite into local scope 
and since the compiler finds a suitable function in this inner scope it 
doesn't bother to search broader scopes (like global scope).

So I think this is OK.

Thanks,

Michael Marcin