[Haskell-cafe] another FFI question

[Niels Aan de Brugh]

On Sat, 2008-06-21 at 00:20 +0400, Bulat Ziganshin wrote:
> 
> probably, you don't understand differences between OOP classes and
> type classes. look at http://haskell.org/haskellwiki/OOP_vs_type_classes
> and papers mentioned there

I took the liberty to update some of the C++ code on that page so that
the code will compile and work as expected.

Reading the rest of the page I have a question regarding section 4.2,
Type classes correspond to parameterized abstract classes (Gabriel Dos
Reis). This is probably more related to C++ than Haskell, so not all
readers might find it of interest.



--- off topic ---

Looking at 4.2, it's interesting to see how type classes are translated
into C++. Admittedly C++ isn't the prettiest language to look at, but I
feel one must be fair, and the code presented at the Wiki is too
complex, using virtuals and templates where they're not needed. Normal
overloading suffices.

Operator definitions can already be placed outside of a class scope,
e.g. the many overloads of << to print something to a stream. By default
these definitions cannot access private members of their operands but
the code at the Wiki has the same problem.

Just a small example:

// --- somewhere in a file ---
struct Aap {
    int x;
    explicit Aap(int x) : x(x) {}
    Aap add(Aap const& other) const { return Aap(x + other.x); }
};

// --- somewhere in another file ---
Aap operator+(Aap const& lhs, Aap const& rhs) {
    return lhs.add(rhs);
}

Overloading functions is not different. I've omitted some other (more
complex) ways you can achieve the same (aside from normal overloads it's
also allowed to introduce new specializations later).

At the time of writing it's not possible in C++ to restrict a type to
allow a certain operation (concepts[1] will hopefully fill that gap some
day, but I think it'll be a long time before tool support is ready for
prime-time). Rather, it's common to just use the operation and let the
compiler generate a (sometimes very cryptic) error message if it cannot
find the right code.
[1]: http://www.generic-programming.org/software/ConceptGCC/

A very poor man's solution to see if an operation is defined would be:

template<typename T> struct has_add;
template<> struct has_add<Aap> {};
int main() {
    (void)sizeof(has_add<Aap>); // okay
    (void)sizeof(has_add<int>); // fails to compile, put useful comment
here
}

It's not hard to add some primitive predicate logic (at compile time) to
resemble type classes in some sense, but I think this is off-topic
enough as it is.

Regards,
Niels