1 of 23

for-in

2 of 23

github.com/bakkot/for-in-exploration

3 of 23

Recap

  • Reflect.ownKeys is ordered
  • for (a in b) {...} is not

4 of 23

5 of 23

6 of 23

Historically there was wild divergence.

7 of 23

There still is for Proxies and

in edge cases.

8 of 23

But there's a broad space

with (necessary) agreement.

9 of 23

Problems with current spec

  • Interpretation for proxies
  • Failure to capture constraints

10 of 23

Goals

  • Nail down behavior for proxies
  • Maybe also common case
  • Clarify spec text in general

​

11 of 23

Non-goals

  • Anything which would require anyone to make changes for non-proxy cases.

​

12 of 23

So.

13 of 23

Proxies

14 of 23

Every engine does something different for proxies.

15 of 23

Uncaptured constraints

​

  • Horizontal order
  • Vertical order
  • New keys are not hit
  • Etc

16 of 23

The common case

​

(way too much text)

17 of 23

The common case

​

  • No proxies or host objects
  • Fixed proto chain
  • No new props up the chain
  • Fixed enumerable attributes
  • Nothing removed then added back
  • Non-enumerable props don't shadow
  • No shadowing props are deleted
  • No proto props are deleted

18 of 23

The common case

​

See associated repo.

19 of 23

Every engine agrees in

the common case.

20 of 23

21 of 23

Make the reference

implementation normative

for proxies.

​

Proposal 1:

​

22 of 23

Make the reference

implementation normative

for the common case.

​

​

Proposal 2:

​

23 of 23