Generalised Ref forms

[kiranandcode]

It would be nice if the following worked:

#lang rhombus

class Example():
  implements Refable
  override method ref(index, ...):
    println([index, ...])

Example()[10,20]

[mflatt]

I think this generalization — allowing multiple dimensions for a ref —could be made to work. When static information is available, the information could be used to make there a suitable number of dimensions are used (i.e., the right number of comma-separated subexpressions within []).

In that case that static information is not available, though, things that are currently rejected statically would turn into dynamic errors. For example, x[1, 2] is currently a syntax error if x is a variable or an expression without any particular static information, but it would be allowed, and then there would be a dynamic error if x is a typical indexable value like a list, array, or map.

So, there's a tradeoff between supporting the generalization and providing better and earlier errors for the common 1-dimensional case. If multi-dimensionsal cases will be rare, then it might be better to leave that to another path, such as a ref method that accepts more arguments or using a list of indices as an index.

I am not immediately sure of the right choice here.

(All of this is about the default #%ref. A variant of Rhombus can always use a different #%ref, either more limited or more general.)

[jackfirth]

I'm not all that into this particular generalization because it doesn't buy much over just writing a[x][y]. It's only one character more.

[mflatt]

Another idea: Rhombus could provide a variant of #%ref (in dynamic and static forms) that allows multiple dimensions. So, in contexts where multiple dimensions are desirable, the extended version of #%ref could be imported. But by default, behavior and reporting for [] refs stays as it is now.