• Simen kjaeraas (40/40) Dec 21 2010 I've noticed that (std.typecons') tuples now are assignable to
• bearophile (10/37) Dec 23 2010 I think it's OK. But the documentation needs to explain about this case.

"Simen kjaeraas" <simen.kjaras gmail.com> writes:

I've noticed that (std.typecons') tuples now are assignable to
structurally equal tuples. I'm not sure I agree fully on the choices
made, and figured a discussion might be in order.

1:
Tuple!(int, "a") a;
Tuple!int b;
b = a;

IMO, this is good and correct. A tuple without field names is to me a
mere black box with data inside, and throwing some other data in there
is ok.


2:
Tuple!( int, "a" ) a;
Tuple!int b;
a = b;

This, I am not so sure about. A black box is turned into structured
data. I guess it's ok.


3:
Tuple!( int, "a" ) a;
Tuple!( int, "b" ) b;
a = b;

This I feel, is wrong. a is of a different type from b.


4:
void foo( Tuple!int );
Tuple!(int, "a") a;
foo( a );

This should work. It may be that alias this is not yet good enough to
make this work, but I believe it should be doable by having tuples with
named fields use alias this to turn into tuples without named fields,
and tuples without named fields taking care of []-lookup.
Basically:

struct Tuple(T...) if(containsStrings!T) {
     Tuple!(noStrings!T) get() {...}
     alias get this;
}

struct Tuple(T...) if(!containsStrings!T) {
     T field;
     alias field this;
}

-- 
Simen

bearophile <bearophileHUGS lycos.com> writes:

Simen kjaeraas:

 1:
 Tuple!(int, "a") a;
 Tuple!int b;
 b = a;
 
 IMO, this is good and correct. A tuple without field names is to me a
 mere black box with data inside, and throwing some other data in there
 is ok.

I agree.


 2:
 Tuple!( int, "a" ) a;
 Tuple!int b;
 a = b;
 
 This, I am not so sure about. A black box is turned into structured
 data. I guess it's ok.

I think it's OK. But the documentation needs to explain about this case.


 3:
 Tuple!( int, "a" ) a;
 Tuple!( int, "b" ) b;
 a = b;
 
 This I feel, is wrong. a is of a different type from b.

I think tuples are meant to be purely structural types, so even adding field
names they are compatibile. It's a design choice.

In Python they have recently added collections.namedtuple, they are tuples with
an optional name:
http://docs.python.org/dev/library/collections.html#collections.namedtuple

If Phobos tuple gain an optional name as I have once suggested, then two tuples
with different names are two different types (so we are back to nominative
typing).


 4:
 void foo( Tuple!int );
 Tuple!(int, "a") a;
 foo( a );
 
 This should work.

If the case 1) works, then this case too has to work.

Bye,
bearophile