Re: [Boost-users] enable-if overloading out the wazoo

Subject: Re: [Boost-users] enable-if overloading out the wazoo
From: Jeremiah Willcock (jewillco_at_[hidden])
Date: 2010-04-23 17:35:54

On Fri, 23 Apr 2010, John Dlugosz wrote:

> I tried something like this:  First, the general form in one header known by everything:
>
>  template <typename T1, typename T2>
>  dim_s foo (const T1& left, const T2& right);
>
> Then, in a header that gets more concrete, provide its own form that is 
> meant to be used when objects of those types are involved:
>
>   template<typename T1, typename T2>
>   typename enable_if_c<
>      is_base_of<genericbase,T1>::value &&
>      is_base_of<genericbase,T2>::value,
>   dim_s>::type foo (const T1& left, const T2& right);
>
> Here, the template defined in this header derives from genericbase, so I 
> have a hook to say "all types derived from any specialization of this 
> template".
>
> Now, this actually causes an ambiguity.  The latter is not a 
> specialization but a peer overloading, and when both are present, both 
> are templates that form exact matches and template arguments at the 
> least restrictive way to refer to the type.
>
> This function is already internal plumbing, so I don't mind it being a 
> little messy to use.  What I'm really doing is multimethod dispatch: 
> various concrete classes, generic versions for making such, and special 
> combinations all have their own efficient logic, with a fallback (slow) 
> that always works.  And I want all the plumbing to disappear at compile 
> time so it efficiently calls the efficient function.  So if I'm really 
> guilty of the "X for Y" problem, that is what I'm really trying to do, 
> and if there are any suggestions on doing that better in general, please 
> let'em rip.

Have you seen <URL:http://www.codeproject.com/KB/recipes/mmcppfcs.aspx>?

> As for the road I'm on now, the simplest thing is to somehow give each 
> form a priority.  I did that by adding a 3rd parameter.  The general 
> form declares a void*, and the specialized form declares an int.  The 
> caller always gives a zero.  I'm wondering what is the longest chain of 
> different priorities I can use with this method, using only compiler 
> primitive types that the optimizer should be able to toss when the 
> function is inlined.  I know I could use dummy classes with any number 
> of derivations, but I'm thinking I can get several levels just using 
> primitive types.  Any suggestions?

What about (... is for you to fill in):

template <..., typename X> // Priority 0
void foo(..., X);

template <..., typename X> // Priority 1
void foo(..., X*);

template <..., typename X> // Priority 2
void foo(..., X**);

Then call foo(..., (void*******************)0) (for however many priority 
levels you want to allow).  Perhaps you can even use a default argument to 
fill in the value in C++0x, but I'm not sure about that.  You can of 
course write a wrapper function that adds in the dummy last argument.

-- Jeremiah Willcock