Re: Unordered setsmaps, for when ordering is hard/expensive/unwanted?

> Tab Atkins Jr. <mailto:jackalmage@gmail.com>
> September 1, 2013 11:38 AM
> [...]
>
> Private correspondence with Mark Miller revealed that I may have not
> been quite as clear as I wanted in describing what I needed.

I've been corresponding with Mark too ;-).
>
> There's nothing wrong with the current "insertion-order" semantics for
> author-level Maps and Sets. Those are fine. However, "insertion
> order" is meaningless for some *UA-provided* Maps and Sets that we're
> producing now or will in future APIs.

It doesn't matter who wrote a map or set in what language. If a spec 
defineds a map or set, we (Mark first, I'm right behind him) want 
determinism. Quoting Mark:

"For spec'ed abstraction X that inserts into a visible table M, if we 
want the spec of abstraction X to be deterministic we need to specify in 
what order it inserts into table M, since that order itself is visible."

> We thus have to define an
> alternative ordering,

That does not follow. We have to define an ordering. Best if it can be 
defined as if the map or set were self-hosted, so insertion order suffices.

> and for some types of content, doing is
> difficult or expensive.

Which types of content?

/be

Received on Sunday, 1 September 2013 21:25:51 UTC