I strongly suggest to "burn with fire", I mean, warn and then deprecate the special case of automatic tuple unpacking in foreach loops in dmd 2.067:


void main() {
    import std.typecons, std.range;

    auto data1 = [tuple("red", 10), tuple("blue", 20)];
    foreach (x, y; data1) {
        pragma(msg, typeof(x), " ", typeof(y));
    }

    auto data2 = only(tuple("red", 10), tuple("blue", 20));
    foreach (x, y; data2) {
        pragma(msg, typeof(x), " ", typeof(y));
    }
}



Output:

uint Tuple!(string, int)
string int


See issues:

https://issues.dlang.org/show_bug.cgi?id=9817
https://issues.dlang.org/show_bug.cgi?id=7361

It causes troubles today because it's a special-cased trap. And it will cause even more troubles in future if/when we add built-in tuple unpacking to D. So please remove this wart, thank you.

Bye,
bearophile

I loved this feature when I first found out about it but after reading your arguments against on the bug tracker I have to agree with your position. It would be much nicer if it worked with uniform tuple syntax like

  foreach (i, {x,y}; [tuple('a',`b`),tuple('c',`d`)])

where i is size_t and x,y is char, string

which is I think what you were proposing.

Tuple unpacking is necessary for things like bypair. Why not unpack consistently ?

deadalnix:

> Tuple unpacking is necessary for things like bypair. Why not unpack consistently ?

I'm all for unpacking consistently, but to reach consistency you first have to break something, the iteration on arrays or the iteration on ranges of tuples. The first is documented and it's present since D1, the second is recent and undocumented and it causes evolution troubles for D toward possible implementations of tuples. What do you want to kill? I prefer to kill the second, and do it yesterday.

Bye,
bearophile