Full std::type_index interface support

[foonathan]

ctti::type_index should have the full interface as std::type_index. Currently only missing the comparison operators.
Also, name() should really return a const char* but this has the null-termination issue discussed in my PR.

[Manu343726]

I'm agree with this. About name() returning ctti::detail::string, that was a try in the line of "ctti::detail::string only", the first alternative I was talking about in your PR.

[Manu343726]

I have been checking docs from both std::type_index and ctti::type_info. Since ctti features have no runtime overhead, there's no point for me in wrapping a type_info to get a type_index, I would just provide features of both at once (What ctti::type_id_t currently does).

That said, type_id_t misses the before() function from type_info, plus comparison operators. Since the collation order is implementation defined, it offers no order guarantees. I was thinking on using hash for comparison in addition to in/equality. Except from MSVC putting "struct" in front of struct type names, our names should be equal on all platforms, yielding same hash, guaranteeing same type ordering.

[foonathan]

Couldn't we remove the prefix to have the same name? Otherwise I agree.

[Manu343726]

Of course we can, but it may be expensive:

if(name.substr<0,8>() == "struct ")
    return name.substr<8,name.size()>();
else
    return name;

[foonathan]

Then pack an #ifdef _MSC_VER around it :D

Maybe you can avoid the branch then completely, is the struct before each type?

[Manu343726]

No, for classes it prepends class, and no prefix is added for basic types. If struct and class had the same length we could use a trait to check for prefix. Also is not that simple since when the type has cv qualifiers struct/class is no longer a prefix.

[foonathan]

Damn, then at least make the check only on MSVC. A hash that doesn't change with the platform would be great.