containers VS ghc-proposals

"Make invalid laziness unrepresentable" means you should use strict versions of container types instead of lazy ones. However, for better or for worse, sometimes the "strict version" of a data type is not actually a strict data type, it's just a strict API to the lazy type. Examples include Data.Map.Strict (not Data.Map or Data.Map.Lazy) or Data.HashMap.Strict (not Data.HashMap.Lazy) (sadly there is no Data.Sequence.Strict but perhaps there will be one day).

Indeed it's rather subtle. See https://github.com/haskell/containers/issues/752 for more discussion on the matter. Nonetheless, I believe it is "spine strict" in the sense that thunks for all values always exist, even if the spine that holds them can be rearranged due to laziness.

Note that the list index operator has O(n) complexity. Ideally you'd want to use something like Seq from the containers package

Edison and containers both have sequence types that support efficient, cons, snoc, viewL, viewR, append, map, and length.

containers

Calling linked lists Haskell's "primary data structure" seems off-base to me. Yes, there's String, yes, there's built-in syntax for List... but there's also everything in containers, and vector is pretty easy to use in practice, though it would probably be good for more learning material to mention it more prominently.

Haskell has much more sophisticated immutable data structures, you can find them in the "containers" package: https://hackage.haskell.org/package/containers

I asked this on the libraries mailing list but thought posting here would bring in potentially more responses. I made a recent change to the behaviour of lookup and find (see here for more details: https://github.com/haskell/containers/pull/800).

Rust has nothing on Haskell. containers, which might as well be considered part of the standard library, has been out for almost 14 years and is still 0.x

> I think Haskell needs a way to graduate (or retire) language extensions

https://github.com/ghc-proposals/ghc-proposals/pull/601

Additionally, while this proposal is focused on the simple implementation and so in the current proposal or patterns do not bind any variables as per #522, if we do get settled on syntax and then later wish to expand to allow binding variables as in #43, then the syntax is again clean and discoverable.

Yeah, a relatively natural thing would be to be able to opt-in to having the HasField instances be available in all contexts. The (not yet implemented, but accepted) "Modifiers" GHC proposal might be nice for that.

The local modules proposal imo is excellent as-is.

Why would you drop TypeFamilies? It seems fairly popular, given that it was just one vote short of getting added to GHC2021. Do you consider it a less-ideal compromise, that is subsumed by your second proposal (Core with GRTT)?

as syntax is contentious a topic as ever, David (the main proposal author) and I would like to invite you to give your input on https://github.com/ghc-proposals/ghc-proposals/pull/585, where we propose to change the syntax of the accepted proposal #522 introducing Or patterns. In particular, we'd like to know

This looks great, congratulations on completing it! I assume you didn’t get an opportunity to evaluate the delimited continuations based eff library? It would’ve been nice to see Cont included as one of the effects covered, though perhaps not available in all the systems you looked at? Maybe some future work.

I do hope qualified exports are available someday: https://github.com/ghc-proposals/ghc-proposals/pull/283

Right. Being able to disable "custom type warnings" at use sites was part of an earlier proposal (https://github.com/ghc-proposals/ghc-proposals/pull/454) but it ended up being shelved.