Re: Proposal: Non-allocating way to iterate over a Data.Map: traverseWithKey_

Hi Ryan,

-----Original message----- From: Ryan Newton Sent: 6 Aug 2013, 09:51

Milan, why does it bother you that there is no specified order? I'm perfectly happy as long as its deterministic on a particular machine.

it seems to me that having specified order (e.g., ascending) is more useful, if it can be provided. If I use traverseWithKey_ to process the elements (e.g., write them to a file, add them to other structure, send across the network), it seems to me that process them in ascending order is more beneficial that iterating over them in some order specified by the shape of the structure. I am not saying that having an operation without defined order is useless, only that operation with ascending order seems more useful to me.

Aren't we already used to using Data.Map.toList and Data.Set.toList where order is not specified?

Well, we also have toAscList and toDescList. Moreover, toList is actually toAscList, so I wonder how many libraries would be affected if toList started returning the elements in some random order :) My point is this -- if we think having an Applicative / Monadic iteration which does not allocate is useful, I believe we should also provide such an iteration in ascending order. Cheers, Milan