Bug 12313 – Unneeded stack temporaries created by tuple foreach

This issue is present for both D1 and D2 but seems to be partially compensated by better optimizer in dmd2. Current foreach implementation (ForeachStatement::semantic) creates a local stack temporary for every element of expression tuple foreach it iterates over: === test.d === struct A { int a, b, c; } void main() { A a; // this empty loop creates variables as copies of every member of a foreach (var; a.tupleof) {} } ====== There are two issues with it: 1) stack space can easily go out of control with algorithms on long tuples (it has just his us in Sociomantic) 2) it prevents one from writing to `var` and forces to use ugly index idiom instead: `foreach (index, var; a.tupleof) { a.tupleof[index] = 42; }` Can't tell possible impact of removing those temporaries right now, needs some experiments.

Have you tried using "ref"? IIRC that worked for my code.

Does not work for D1: "Error: no storage class for value var" It does not also make any sense from language semantics point of view as a.tupleof is effectively tuple of variable aliases. And there is no such thing as `ref alias`. Sounds more like a hack to fix another hack.

(In reply to comment #2) > It does not also make any sense from language semantics point of view as > a.tupleof is effectively tuple of variable aliases. And there is no such thing > as `ref alias`. > > Sounds more like a hack to fix another hack. I see it more like the foreach body being like a function (whose body is implicitly inlined). In this context, "ref" works out fine. Since .tupleof only enumerates fields, and not functions, each member can be mapped to a concrete value (and thus can work with "ref").

foreach != tuple foreach latter is more like template declaration that gets instantiated for every tuple element as an argument

(In reply to comment #4) > foreach != tuple foreach > > latter is more like template declaration that gets instantiated for every tuple > element as an argument If you want to iterate the elements of the container (array, range, tuple, etc) without copying, use 'ref'.

D1 does not have ref foreach. tuple is not a container.

Since D2 has a trivial workaround, can we agree that this is a D1-only bug?

I'd say it is D1 bug and D2 enhancement request ;) But for practical concerns it can be considered D1-only, no problem.

Commit pushed to dmd-1.x at https://github.com/D-Programming-Language/dmd https://github.com/D-Programming-Language/dmd/commit/6ed4a173f41b7c8dc6f024ae11189b522213a585 fix Issue 12313 - Unneeded stack temporaries created by tuple foreach

Fixed by implementing D2-style foreach ref for tuples.

Thanks, this is really a big help!