Issue 4999 - Add Kenji Hara's adaptTo() to Phobos

[9rnsr]

http://d.puremagic.com/issues/show_bug.cgi?id=4999

Changes:

• Add structuralCast and structuralDownCast in std.typecons module
• Remove unused and undocumented structuralCast which was in std.exception.

[ghost]

Could you indent predicates so they're like this:

template T()
    if (...)
{
}

It makes it easier to distinguish a template/function definition to a regular if statement block.

[DmitryOlshansky]

I like the functionality but structuralCast ?!
I thought it was all about casting structs having the same field structure and then turns out it's about adapting 2 unrelated clases/interfaces having the same sets of functions. I'd call it duckCast !!!

[9rnsr]

My attitude for the indentation of template constraint part is not determined. In past days, I used the same style that you suggested (std.conv and std.format is mostly based on the style). But after a while, I noticed that the style has the problem with complex template constraints.

template T()
    if (... && ... &&
        ... && ...)    // 8 indentation
{
}

If the constraint needs more than one line, the second line needs 8 space indentation. I feel it is "too far" from the left edge.
Again, I don't have strong argument for the style. But it is not yet determined in the "Phobos style guide". So I'd like to keep this.

[9rnsr]

I like the functionality but structuralCast ?!

Originally that was named adaptTo. I renamed it to structuralCast based on the Andrei's suggestion.
Better name is welcome!

[ghost]

What does this code do? It looks really weird.

[9rnsr]

This makes a function type FX that has a return type R and parameter types P0.

[dnadlinger]

C function type syntax. Eww…

[alexrp]

I'm pretty sure that's slated for deprecation, no?

[ghost]

Yeah I think it will be. Seems like alias FX = R function(P0); would do here.

[9rnsr]

In here, FX should be a function type, not function pointer type. So, alias FX = R function(P0); is not correct.

[dnadlinger]

Couldn't you just use FunctionTypeOf!(R function(P0)) or something like that? I think that would make the code much more readable.

[9rnsr]

It's good idea. Fixed.

[sgraf812]

Concerning names: How about

Orig obj = ...;
auto ifc = wrapAs!Interface(obj); // or
auto ifc = obj.wrapAs!Interface; 
// and
obj = unwrapTo!Orig(ifc); // or
obj = ifc.unwrapTo!Orig;

and extend it to work on structs? Haven't tried it, just based on what David wrote.
It would merely involve an alias-this in a templated class.

[andralex]

I think we need short, memorable, single-word names like wrap and unwrap.

[DmitryOlshansky]

I think we need short, memorable, single-word names like wrap and unwrap.

+1

[9rnsr]

Renamed function names to wrap/unwrap.

[andralex]

Great. We'd need some links and expanded documentation and examples. (One simple reference would be e.g. http://en.wikipedia.org/wiki/Structural_type_system).

[andralex]

This will be a very influential piece of work, thanks @9rnsr, There are quite a few ways in which we can expand on this idea, will flesh them out later.

[ghost]

I'm a little confused about this, since I thought wrap was supposed to be generic and allow code such as:

interface Interface
{
    int foo();
    int bar(int);
}

struct Pluggable  // note that it's a struct
{
    int opDispatch(string name, A...)(A args) { return 100; }
}

Interface i = wrap!Interface(Pluggable());

This currently errors inside of phobos with:

std\typecons.d(2875): Error: template std.typecons.dynamicCast does not match any function template declaration. Candidates are:

If the wrap function is only supposed to be used with objects it should be constrained to disallow non-class types, and it should also be properly documented.

[9rnsr]

@AndrejMitrovic It's a current implementation limitation which I had not recognized. For unwrap operation, current implementation uses the Objecttype as a most generic base type. However it won't work for any struct types.
We should relax the limitation in the near future release.

[ghost]

Ok, great to hear.