Re: [boost] Review Request: impl_ptr (pimpl)

On 08/23/2017 07:07 AM, Seth via Boost wrote:

On 21-08-17 11:48, Vladimir Batov via Boost wrote:

...

That's ugly and wrong. /I'm just gonna say it's neither ugly nor wrong./

OK, OK. :-) We certainly won't be arguing that. :-) I just thought that the words themselves clearly indicated that it was in my view, by my taste. I do not feel comfortable seeing the code that looks like the user introduces something, fiddles in the boost (or std) namespace. I have a feeling I might be not the only one feeling that way. :-)

On 22-08-17 09:50, Gevorg Voskanyan via Boost wrote:

...

so for a long time now, we have a precedent too.

Gevorg was certainly correct pointing out this boost::assertion_failed() precedent... Something I have in my code. :-)

One of a great many.

Same for specializations of std::hash, boost::hash, boost::fusion::extension::*, boost::spirit::traits::*, etc.

That said, I'm not convinced there isn't a "better" way: why not implement another layer of indirection. Make it a trait:

boost::impl_ptr::traits::implementation_for::type

could be made to refer to the user-defined implementation type. That way the "pain" is only in 1 line where the trait is specialized.

Before we embark on exploring other possibilities, is there something in particular that you object to in the proposed template<> struct boost::impl_ptr<Foo>::implementation declaration?