Unintuitive priority of or?

[vcunat]

A while ago I was taken unawares by the fact that the operator priority is higher for or than for function application. That is strognly against my functional-programming intuition of application always binding the strongest. Example of the current state:

{ truth =
  set.attribute or function param == (set.attribute or function) param;
}

I'm aware that changing such things is not to be taken lightly – maybe in nix-1.10 it might be bearable? (Unless you actually prefer the current priority.)

[vcunat]

Perhaps there are some good arguments for that decision. One can't find such information in the nix manual... (but the priority order is documented in there clearly)

[vcunat]

I don't think there's sufficient motivation to keep an open ticket for this 🤷

[zimbatm]

It belongs on a "language quirks" inventor page :-)

[vcunat]

😄 the formulation made me recall http://lambda-the-ultimate.org

[roberth]

I think we could make the parentheses in the third operand semi-mandatory, by emitting a warning if the . or ast node is unparenthesized and occurs in a function application. This check would ideally be done during the parsing, because the ast is too lean (and we probably want to keep it that way).
It's not perfect, but at least we can fix the confusion without breaking compatibility.

ie don't warn for the unambiguous case:

(set.attribute or function) param

EDIT: at first I thought this looked unambiguous, but I've changed my mind:

set.attribute or (function) param

Perhaps there are some good arguments for that decision.

Fwiw, there's no or operator; only the _ . _ or _ ternary operator, with the same priority as _ . _.

[vcunat]

Sounds like a good idea, from UX perspective at least. I think it will be rare to use or with a function on the right side, so warn-forcing explicit parenthesis in that case should be a low price overall.