[RFC] Array spread syntax

[zth]

Array spreads are quite common in JS land, and they're pretty awkward to do right now in ReScript. You essentially need to do this (using Core):

let arr1 = [1, 2]
let arr2 = [3, 4]
let arr3 = [5, 6]

let arr = Array.concatMany([7, 8], [arr1, arr2, arr3])
let arrOnlySpreads = Array.concatMany([], [arr1, arr2, arr3])

Which compiles to:

var arr1 = [
  1,
  2
];

var arr2 = [
  3,
  4
];

var arr3 = [
  5,
  6
];

var arr = [
    7,
    8
  ].concat(arr1, arr2, arr3);

var arrOnlySpreads = [].concat(arr1, arr2, arr3);

Compiling to concat is fine imo, but it'd be great if the ReScript above could be written like this instead:

let arr1 = [1, 2]
let arr2 = [3, 4]
let arr3 = [5, 6]

let arr = [7, 8, ...arr1, ...arr2, ...arr3]
let arrOnlySpreads = [...arr1, ...arr2, ...arr3]

This could parse to the first ReScript snippet using concatMany, that's fine.

I believe this would be a pretty harmless change (as of now you need to write the exact code it parses to manually) and being a significant gain in ergonomics. Remains to be seen how difficult it is technically (need to account for things like [1, ...arr1, 2, 3,4 ...arr2] etc).

Thoughts?

[zth]

Side note, but concat seems to be pretty slow compared to the native array spread. Something we can consider for v12 or when we want to start allowing ES5+ JS.

https://jsbenchmark.com/#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

[TheSpyder]

concat seems to be pretty slow compared to the native array spread

Don’t trust micro benchmarks. The performance at scales of “two numbers per array” is so fast either way that it’s irrelevant. With larger arrays spread will naturally be slower because it’s iterating, not just copying.

One example I found without much research:
https://jonlinnell.co.uk/articles/spread-operator-performance

[zth]

@TheSpyder that's great! 😄 Sounds like that's not something worth worrying about then.

[cometkim]

As the list{} syntax already supports this, it's good to have for all the first-class data literals we have, including #6545 and #6181