Should treat const and enum (at least const enum) value the same as literals for indexers.

[vilicvane]

const DATA_END_INDEX = 2;

enum OptimizedDataIndex {
    start,
    data,
    end
}

interface OptimizedData {
    0: number;
    1: any[];
    2: number;
}

var data: OptimizedData;
var start = OptimizedData[OptimizedDataIndex.start];
var end = OptimizedData[DATA_END_INDEX];
// expecting no error. `start` and `end` should be numbers.

This should also apply to something like tuples directly written as [TypeA, TypeB, ...].

[danquirk]

Presumably your example intended to index into data and not the interface type. This is sort of the simplest case of constant propogation/flow control since the compiler can at least assume the name<->value pairs here are immutable.

[vilicvane]

So can I take that as "this suggestion is reasonable"? :D

[RyanCavanaugh]

Summary: When we see a property access where the indexing expression is an enum member, we should resolve the property access as if the indexing expression were actually the computed value of the enum member's value.

In other words, var x = Foo[E.bar] is equivalent to var x = Foo[2] if we can resolve E.bar to its value 2.

[vilicvane]

And also consts, I think.

[jbondc]

Added a patch for the const enum access. Would also like to see these work:

const one = 1;
module dot {
   export const one = 1; 
}
const enum numbers {
    one = 1
}
const oneRef = one;

interface OptimizedData {
    0: number;
    1: any[];
    2: number;
}
let data:OptimizedData 

// works
data[1] // any[]
data[numbers.one] // any[]

// should work
data[one] // any[]
data[dot.one] // any[]
data[oneRef] // any[]

The last 3 could work but require better tracking of constants and constant propagation like @danquirk mentioned. Don't think you need flow control to do it properly, have gotten it working without anyways.

[RyanCavanaugh]

Approved for const enum members only. @jbondc can you get your PR up-to-date?

[mhegazy]

@jbondc, we would appreciate it if you can keep it only for const enums, and not for generic constant propagation.

[mhegazy]

thanks @jbondc!