This is a well written bug fix, with tests. However, existing code relying on the numeric sequential keys of the output of this function will break, if we apply this fix.

@acoulton could you kindly comment what to do in this case?

Tricky. Since it changes the response, I think it would have to go to 3.4 with a clear note in the upgrade guide.

It will only be if end-user code is specifically looking at the keys, I think, that it's an issue - they'll still come out in the same sequence as far as I can see? It's probably quite an edge case, but still not safe for 3.3.

@tommcdo if you disagree we'd welcome your opinion.

I think it's quite unlikely that users are currently relying on the numerical keys with the current implementation, since they're quite frankly not useful at all. However, this is not strictly a bug fix, so it's perfectly reasonable to hold off until 3.4.

However, this is not strictly a bug fix, so it's perfectly reasonable to hold off until 3.4.

@tommcdo could you please PR this to 3.4/develop branch.

What we do with this? to 4.0.0? Looks good to me.

I suggest a contrib manually merges this PR to the 4.0 branch and adds to upgrade notes