Faster iteration experiment

[mweststrate]

Object.keys is faster than Reflect.keys. Can we leverage this somehow? (at the cost of symbol props etc) https://www.measurethat.net/Benchmarks/Show/18717/0/reflectownkeys-vs-objectkeys-preformence#latest_results_block

[markerikson]

FWIW, I've actually been doing some RTK perf optimizations over in reduxjs/redux-toolkit#4361 (comment) , and found that 1) it was faster to do some reads on a const ids = current(state.ids) array instead of a wrapped array, 2) the majority of the remaining overhead seems to be in Immer, 3) a lot of it looks like it's in finalizeProperties

[mweststrate]

See also #1107. There are several ways to make Immer quite a bit faster, but it would required much more loosely treating things that have getters/ non-enumerables / symbolic properties, which would benefit the majority of the user base, but also create enough failure scenarios to flood the issue tracker😅

[markerikson]

@mweststrate out of curiosity, how much of a perf difference do you think this will make?

FWIW it looks like only one test failed:

FAIL __tests__/base.js
  ● Test suite failed to run

    TypeError: Cannot perform 'get' on a proxy that has been revoked

      2379 | 		})
      2380 |
    > 2381 | 		expect(state[sym].x).toBe(2)
           | 		                  ^
      2382 | 	})
      2383 | }
      2384 |

      at __tests__/base.js:2381:21
      at testObjectTypes (__tests__/base.js:2368:2)
      at __tests__/base.js:1861:28
      at __tests__/base.js:1860:3
      at runBaseTest (__tests__/base.js:61:2)
      at Object.<anonymous> (__tests__/base.js:32:4)