Add filterM

[treeowl]

Thus far, this proposal has gotten good reception on the libraries list.

[tibbe]

Have you benchmarked and showed that it's faster than filterM . toList. In the case of bytestring we rejected adding a patch like this because using rewrite rules we got the same performance without growing the API.

[treeowl]

Oh, you mean filterMSeq f = fmap fromList . filterM f . toList? I guess that's an option, and probably a good implementation, but it's a lot less convenient. There seems to be a fairly broad agreement that making it more convenient to use Data.Sequence is a good thing. However, that implementation does reveal a nice property: filterMSeq is fundamentally an applicative filter producing a Seq; it's perfectly happy to take any Foldable as input.

[tibbe]

The bigger question here is: should we add applicative versions of every function to the containers API? There's an argument to be made for it. You might want to perform some effect when doing e.g. an alter. The obvious downside is huge amount of code duplication and a much harder to understand (and less composable) API. That's why we've been pushing back on this so far. Our functional container APIs are already much bigger than their imperative siblings. I'm worried we'll end up with APIs so big that their hard to use.

[treeowl]

There's been a big trend to move away from lists as the only container type that gets a lot of good functions, and I personally think that's a good thing. Making people roll their own functions to use containers makes them more likely to just give up and use lists or vectors instead, whether or not those are the most appropriate. It is true that the API for IntMap is a bit intimidating (what, for example, can updateWithKey do that update cannot, aside from weird things with broken Eq instances?) but Seq is another story.

[m-renaud]

Appologies if this is too tangential: I haven't seen a proposal but is there any reason why filterM from base couldn't be changed to:

filterM :: (Applicative m, Foldable t) => (a -> m Bool) -> t a -> m [a]

Then not every data structure would have to provide its own implementation of these functions since it would just use the Foldable instance, which Sequence already has. This is obviously a much larger change, but was wondering if this has been considered.

Edit: duh, this doesn't have the same type as the filterM in the PR.