"John Harrison" <> wrote in message
news:c6bbir$a4ii7$...
>
> "Conrad Weyns" <> wrote in message
> news:c6b9ut$1ece$...
> > vc7.1 has no problem with the following snippet:
> >
> > template <typename T> class TTest
> > {
> > std::list<T*>::iterator m_It;
> > };
> >
> > but metrowerks codewarrior 9.2 kicks my butt and online Comeau confirms:
> >
> > 1. MW:
> > Warning : 'typename' is missing in template argument dependent qualified
> > type
> > Tests.cxx line 52 std::list<typename T*>::iterator it;
> >
> > 2. Comeau:
> > "ComeauTest.c", line 5: error: nontype "std::list<_Tp, _Alloc>::iterator
> > [with
> > _Tp=T *, _Alloc=std::allocator<T *>]" is not a type name
> > std::list<T*>::iterator m_It;
> > ^
> >
> >
> > There were not that many places to insert "typename", so after a while I
> > found that the following makes
> > CW and Comeau happy:
> >
> > template <typename T> class TTest
> > {
> > typename std::list<T*>::iterator m_It;
> > };
> >
> >
> > Can anyone here help me understand this?
> > regards,
> > Conrad Weyns
> >
>
> When the compiler first looks at TTest cannot tell if iterator is a
typedef
> or a static member in the class std::list<T*>. This is because it does not
> know the type of T at this point. It cannot even find this out by looking
up
> the definition of std::list because there maybe specialisation of
std::list
> which are not visible while it is first looking at TTest (actually that
not
> likely to be true for std::list, but it's true of templates in general).
>
> The classic example of why the compiler needs to know if a 'dependant
name'
> is a typedef or a static member is this
>
> template <typename T>
> class Awkward
> {
> T::some_name * global;
> };
>
> Now is that a declaration of a variable called global of type
T::some_name,
> or is it a multiplication of the static member T::some_name and the global
> variable called global?
>
> In general the compiler cannot resolve these ambiguities so you have to do
> yourself. Nevertheless in most cases the compiler can work out what you
mean
> (or at least take a reasonable guess) so many compilers let these things
go.
>
> Actually this is the simplest of the syntactic ambiguities introduced by
> templates, but I'll not post any others (unless you're particularly
> interested) since I'd have to look them up myself.
>
> John
Many thanks, this has been usefull. I know what to play with now to get a
better mental picture of this.
Conrad
>
>
|
Similar Threads
