Strict maps

[treeowl]

Closes #60

[konsumlamm]

Why no map', traverse', traverseU'?

Also, do we want strict adjustMin etc.? It would be a lot of new functions for something that can easily be done by hand.

[treeowl]

I don't know how many more functions to add. Do we want to do a whole Strict module like containers?

[treeowl]

One option would be Strict modules that only export the strict functions, and not the rest of the API. That might be less confusing than the way containers does it....

[treeowl]

Why no map', traverse', traverseU'?

Also, do we want strict adjustMin etc.? It would be a lot of new functions for something that can easily be done by hand.

For adjusting the minimum, we can't do better than

adjustMinWithKey' f = getSolo . adjustMinWithKey ((Solo $!) . f)

Doing it "manually" with the constructors produces exactly the same Core, unfolding, inlining guidance, etc. So arguably we don't need adjustMin' and such, but I wouldn't be opposed to having them for user friendliness.

[treeowl]

Also, do we want strict adjustMin etc.? It would be a lot of new functions for something that can easily be done by hand.

I just realized I don't understand your last sentence. Implementing adjustMin' using Solo is easy for someone with enough experience, but a lot of people don't find that sort of thing obvious. I don't think it's possible to implement the strict mapping and traversal functions optimally without access to the internals. The only remaining things I can think of are indeed pretty trivial: strict insertion and strict fromList.

[konsumlamm]

I don't know how many more functions to add. Do we want to do a whole Strict module like containers?

One option would be Strict modules that only export the strict functions, and not the rest of the API. That might be less confusing than the way containers does it....

That might be a good idea, given the amount of strict functions. I agree that such a module should only export the strict functions. I'm also used to the strict version just having a ' suffix though, idk how others would feel about that. Perhaps we should ask for feedback on that (e.g. on the Haskell Discourse or Reddit).

I just realized I don't understand your last sentence. Implementing adjustMin' using Solo is easy for someone with enough experience, but a lot of people don't find that sort of thing obvious. I don't think it's possible to implement the strict mapping and traversal functions optimally without access to the internals.

Yeah, we definitely shouldn't expect everyone to know that you can use Solo here. I meant using minViewWithKey and insert (or the new pattern synonyms), but that wouldn't be O(1).

The only remaining things I can think of are indeed pretty trivial: strict insertion and strict fromList.

For those I'd say let the user force the elements.

[treeowl]

I'll ask around and see what people think. If we go with a separate module, strict insertion wouldn't be a big deal to include. Even strict extraction could make sense in that context—I wish containers had done that, but it's too late to change now.

minViewWithKey q >>= \((k, v :: Int), q') ->
  Just (... [v + 1])

will produce a thunk to calculate v+1, whereas with

minViewWithKey' q = case minViewWithKey q of
  r@(Just ((_, !_), _)) -> r
  Nothing -> Nothing

this will not.

minViewWithKey' q >>= \((k, v :: Int), q') ->
  Just (... [v + 1])

[konsumlamm]

I opened a discussion on Discourse: https://discourse.haskell.org/t/organization-of-strict-functions/6354.