On Tuesday, 28 May 2013 at 00:29:04 UTC, bearophile wrote:
std.typecons.Tuple supports "structural assignment" before the
change.
The code also works with 2.062.
I know it's not a regression. But you say:
"Named-field tuple should be a subtype of unnamed-field tuple."
You can have sub-typing, or you can have structural typing, but
mixing the two "silently" seems a mess.
Also in your test cases of that commit fdcaba7226c... there is:
+ Tuple!(int, int)[] arr;
+ arr ~= tuple(10, 20); // OK
+ arr ~= Tuple!(int, "x", int, "y")(10, 20); // NG -> OK
If I try to do the opposite, that is to append a tuple(10, 20)
to an array of tuples with named fields, I get an error:
import std.typecons;
void main() {
Tuple!(int, "x", int, "y")[] arr;
arr ~= tuple(10, 20); // Error.
}
So here there is a true sub-typing. So sometimes tuples with
named fields are sub-types and in other situations they act in
a structural typing way. I think this mix of behaviours is a
little confusing.
Basically Tuple would follow structural typing. Even if it is
impossible, it would follow true subtyping right now. That's the
purpose of my change.
Kenji Hara
