dolfin team mailing list archive

[Martin Sandve Alnæs <martinal@xxxxxxxxx>]

On 31 August 2011 21:13, Anders Logg <logg@xxxxxxxxx> wrote:
> On Wed, Aug 31, 2011 at 09:10:21PM +0200, Martin Sandve Alnæs wrote:
>> On 31 August 2011 16:12, Anders Logg <logg@xxxxxxxxx> wrote:
>> > I want to declare the following function:
>> >
>> >  template<class S, class T>
>> >  void apply_markers(S<T>& sub_domains, T sub_domain) const;
>> >
>> > S is a template class (either MeshFunction or MeshMarkers) and T is a
>> > primitive type.
>> >
>> > The above declaration does not work. How should it be declared? I've
>> > tried various other combinations without success.
>>
>>
>> If you can make T a typedef in S, this works:
>>
>>
>> template<typename V>
>> class Foo
>> {
>> public:
>>     typedef V value_type;
>> };
>>
>> template<typename S>
>> void apply_markers(S & sub_domains, typename S::value_type sub_domain)
>> {
>>   S s;
>> }
>>
>> int main(int argc, char *argv[])
>> {
>>     Foo<int> a;
>>     apply_markers< Foo<int> >(a, 3);
>>     return 0;
>> }
>
> Thanks. I think I just found another solution (remains to be seen,
> currently compiling...). I simply do
>
> template<class S, class T>
> void apply_markers(S& sub_domains, T sub_domain) const;
>
> S happens to be a template type over T, but it's not necessary to
> write that out explicitly in the above function definition.

Adding something like "typedef T value_type;" in S makes
this more explicit, and is the standard approach in stl and boost.

Note that "typename" is supposedly considered nicer than "class"
in template<> by modern C++ people. Because T doesn't have to
be a class, but it has to be a typename. No technical difference though.

Also, the reason for the "typename" before S::value_type in my
suggestion is that the compiler can't know whether S::value_type
is a type or variable when parsing the code.

Martin