make empty mutable structs non-singletons

[carnaval]

note to self

[vtjnash]

:)

I think there may even be a few places already where we do so, but inconsistently. Alternatively, you could make mutable singletons unique values instead of being strangely almost equivalent to their immutable cousins.

[eschnett]

Yes, having mutable singletons be unique would be cool. Currently I'm adding an unused Int field to a mutable type for this (is there a better way)?

[yuyichao]

I thought this is only about codegen calling convention rather than object construction? "mutable" singletons should already be unique. What does adding a Int field accomplish?

[eschnett]

@yuyichao

type N end
N()===N()
true

type N1 i::Int; N1()=new() end
N1()===N1()
false

Each call to N1() creates a new, unique object that can be used as abstract id. (An alternative would be using a global counter.) Calls to an empty type constructor such as N() don't; all objects of type N are identical, as if they were immutable.

[yuyichao]

OK, you mean unique as each of them are different, not that it has a global unique value.....

[toivoh]

@eschnett: You can use a field of eg type Void instead of an Int field, should save some space?

But I agree that it would be nice if there could be multiple instances of mutable types without fields. The current behavior was very handy before we had immutables, but now that we do, it just feels inconsistent.

[stevengj]

This came up on discourse recently. I agree that having mutable singletons construct unique (not ===) objects makes sense nowadays.

[stevengj]

Marking as triage since it seems like this should really get decided before 1.0.

[Keno]

Triage says yes, let's make empty mutable structs give non-singleton.