Re: [Boost-users] [multi_index] Got warnings in VC7.1

From: JOAQUIN LOPEZ MU?Z (joaquin_at_[hidden])
Date: 2006-10-10 17:47:02

Hello Shenlei

----- Mensaje original -----
De: ShenLei <sweetmelon_at_[hidden]>
Fecha: Martes, Octubre 10, 2006 11:16 pm
Asunto: [Boost-users] [multi_index]Got warnings in VC7.1
Para: boost-users_at_[hidden]

> Hi,
>   I used multi_index in the simple way:  int<->int unique 
> mapping. But in VC7.1, warning level 4, I got many warnings for: 
> \boost_1_33_1\boost\multi_index\detail\ord_index_node.hpp(453): 
> warning C4510: 'boost::multi_index::detail::
> ordered_index_node<Super>' : default constructor could not be
> generated.
> 
> I have read the doc of multi_index, and tried to turn off "\Gm", 
> but the problem is still. When compiled it in VC8.0 under
> warning level 4, these warnings disappeared. It seems that, the
> template support in VC7.1 is not so good. My test code is listed
> in the end of this article. How to get rid of them? (...>_<...
> The webpage argues me about top-posting, so I cannot put them 
> on).

(Well, you seem to have forgotten to attach the test code,
but I think I don't need it to answer.)

These C4510 warnings are hopefully avoided in what will become
Boost 1.34. You might want to check by downloading a snapshot
of the current CVS at: 

http://www.boost-consulting.com/boost.tar.bz2

Does this solve your problem? Anyway, particularly bothering
(and totally uninformative) warnings like this can always
be shut up using a #pragma warning(disable:xxx).

> Another thing is about compile time. If multi_index is employed, 
> the compile time will be greatly increased (slower 10 times in
> VC7.1, 3 times  in VC8.0).

Yep, Boost.MultiIndex subjects compilers to severe stress :(
although your figures seem to indicate they're getting much
better at compiling metaprogramming stuff.
Have you tried turning precompiled headers on? This usually
has a dramatic (positive) impact on compile times.

> Furthermore, the message generated by compiler is extremely long. 
> I usually get lost in it. ^_^ Is there any way to simplify them?
> Such as using struct T:S rather than typedef in multi_index
> library?

Yes, there's a tecnhique called type hiding which can be used
here. It is explained at:

http://boost-
consulting.com/boost/libs/multi_index/doc/compiler_specifics.html#type_
hiding

> Finally, I have a question on the design of
> template<class Class,typename Type,Type Class::*PtrToMember>
> struct member; In practice, the user of "member" should provide
> the SAME type of PtrToMember to the template arguments
> (Class & Type). If feeded with some wrong types, errors message
> for the hell will be sent out. ^_^ Is there any way to improve 
> it? Such as simply member<T* PtrToMember>, and let the compiler to 
> deduce the type of this memberpointer?

As you point out, *seems* like the compiler ought to be able
to deduce the type without the user explicitly providing it,
but unfortunately this is not possible with current C++ mechanisms.
Once we've got typeof in a couple of years some techniques can
be devised so that something like the terser member<T* PtrToMember>
can be used, but in the meantime we're stuck with the current
syntax.

> Thank you.

Thank you for using Boost.MultiIndex. Hope this helps, come back
if you have further problems.

> ShenLei

Joaquín M López Muñoz
Telefónica, Investigación y Desarrollo