RE: [boost] Re: [type-traits] Patch to alignment_of.hpp for Sun c ompiler

From: Aleksey Gurtovoy (agurtovoy_at_[hidden])
Date: 2003-05-07 00:54:36

David Abrahams wrote:
> Aleksey Gurtovoy <agurtovoy_at_[hidden]> writes:
> 
> > David Abrahams wrote:
> >> > Looks good. I've applied it.
> >> 
> >> Maybe we should be treating Sun in the config as though it doesn't
> >> support in-class static constant initialization (so
> >> BOOST_STATIC_CONSTANT generates an enum)?  It seems like trying to
> >> track down all these cases piecemeal is going to be a 
> losing battle.
> >
> > But then Sun is right to complain there, isn't it? It's true that 
> > there are a lot of places where we are using BOOST_STATIC_CONSTANT 
> > without the corresponding and appropriately guarded definition, but 
> > if we adopt the strategy you suggest, in the end, as more and more 
> > compilers will enforce the rule, we will end up with enums 
> > everywhere; which kind of defeats the purpose of the macro - if I 
> > thought the enums are the way to go, I would never use 
> > BOOST_STATIC_CONSTANT in the first place.
> 
> I was under the impression that a DR had been approved which clarifies
> the point so that if the address of the constant is never needed, no
> definition is needed either. 

Are we talking about
http://anubis.dkuug.dk/jtc1/sc22/wg21/docs/cwg_defects.html#48?
'Cause while the stated intent of the issue is indeed formulated in these
terms, the resolution itself is anywhere but close to it:

"An expression is potentially evaluated unless it appears where an 
integral constant expression is required (see 5.19  expr.const ), is the 
operand of the sizeof operator (5.3.3  expr.sizeof ), or is the operand 
of the typeid operator and the expression does not designate an lvalue 
of polymorphic class type (5.2.8  expr.typeid )."

For a library writer, it pretty much means that if a static integral 
constant member is publicly exposed, it needs to have a definition.
The library writer _herself_ needs it for the tests, if she wants to 
check the value at runtime.

> If I'm mistaken about that, then I'd be inclined to agree with you.

Well, having formulated the above, I just realized that the case is 
question doesn't in fact fall into that category. 'alignment_of_impl' 
is an implementation helper which 'value' member is never used, and,
more importantly, cannot potentially be used anywhere but in the 
context where an integral constant expression is required. 

So, if Sun wants a definition for it, it's definitely a non-conforming 
behavior, and should be either guarded with a BOOST_WORKAROUND #ifdef, 
or indeed by defining BOOST_NO_INCLASS_MEMBER_INITIALIZATION. The 
latter would definitely make our life easier, if it works, and is no
worse for the vendor's testing, if they are to pursue the conformance -
in that case they are supposed to replace the config anyway.

Aleksey