Collection method for associative, but not commutative, reductions about rayon HOT 8 CLOSED

@cuviper well-- consistency is a lot to ask =)

In any case, I feel all right about just requiring associativity.

from rayon.

I don't know if it's promised, but the current implementation should actually preserve order in the reductions, just with associative differences as you note. This is a natural implication of the way the jobs are split recursively and then joined as it returns.

from rayon.

I believe that in https://github.com/nikomatsakis/rayon/pull/120 I went ahead and promised that the order would be deterministic when doing reduce or fold.

from rayon.

Hmm, the new reduce says just associative, but reduce_with still says commutative too. Then sum, mul, min, and max all mention commutative operators.

from rayon.

Are there tests for just-associative reduction functions (String::push?) already?

from rayon.

Using collect into any of the ordered collections will exercise associative non-commutative reduce. Look at their FromParallelIterator implementations.

Any further question about this? Otherwise, I think we can close this issue.

from rayon.

Yeah, the issue can likely be closed. I was more wondering whether there were tests to this effect rather than "well, the implementation works that way".

from rayon.

Ok, I don't think we have any plain mathematical tests of this, but all of the collection and string stuff mandates that it works this way. Also, find_first et al. would make no sense if we didn't maintain order.

from rayon.