Re: [Boost-users] BOOST_STATIC_ASSERT with intel compiler at class scope

From: John Maddock (john_at_[hidden])
Date: 2004-10-14 05:20:25

> The BOOST_STATIC_ASSERT documentation indicates I can use it at template
> class scope (private recommended) - the assert shouldn't get triggered
> unless the class is instantiated.  This seems to work ok with gcc-3.*
> but not with the intel compiler (8.1).
>
> // begin example program
> #include "boost/static_assert.hpp"
>
> template<class T> struct foo
> {
> foo (int) { }
> private:
> BOOST_STATIC_ASSERT (false);
> };

This is a recurring problem, and one for which I should update the 
documentation: if the contents of the assert are not dependent upon a 
template parameter, then a "sufficiently clever" compiler can evaluate the 
assert at the point it is declared, realise that there is no instantiation 
of that template that could ever be legal, and reject the code.  Normally 
this is a "good thing", but unfortunately messes up static asserts as you 
have found.  If you really want an assert that always fails then try:

template<class T> struct foo
{
foo (int) { }
private:
BOOST_STATIC_ASSERT (sizeof(T) == 0);
};

although there is still some question as to whether a "sufficiently clever" 
compiler could regard this as ill formed as well :-(  Fortunately none do at 
present.

John.