haskell.nix VS polysemy

I'm not a great haskeller but I found haskell.nix better for Haskell projects, like the commenter on Discourse suggested. I've had a few issued regarding package versions with nixpkgs that haskell.nix solved.

FWIW, there's also libraries like haskell.nix that solve the caching problem.

I would try haskell.nix.

If you want to hurt your brain, check out haskell.nix. That's some good stuff right there ^^

The haskell.nix framework is good for this.

One cool thing is that this will enable GHC builds using ucrt instead of vscrt in the future. Concretely, together with NixOS/nixpkgs#171418 and its follow-up NixOS/nixpkgs#173498), this will e.g. allow haskell.nix to upgrade to a newer wine pin for TH cross compilation: https://github.com/input-output-hk/haskell.nix/blob/dd13e822529ae5342494969bce8a457522a60100/overlays/wine.nix

I've been using IOHK's alternative infrastructure for this reason. It has its quirks but I've been happier with it. Before that I think I was using developPackage from the nixpkgs haskell tooling which had some introspection ability. You may consider trying that out. But as I remember this will not abide by your version bounds.

There's a lot to like here. Alas, despite minimal dependencies, terminfo is somehow uniquely problematic in haskell.nix.

If you are somewhat comfortable with nix: https://github.com/input-output-hk/haskell.nix supports GHCJS 8.10.x (in particular 8.10.7).

Thirdly, composing arbitrary effects without losing state is really, really difficult. Things are fine when you limit yourself to State and Reader, sure, but once you start with nondeterminism you’ll discover it’s shockingly easy to produce behaviors that are baffling unless you’ve spent a preposterous amount of time thinking about this stuff. (I’ve been bitten in prod by silent state-dropping bugs, and rarely have I been more flummoxed.) Consider this example, which produces silent changes in the semantics of <|> depending on whether you use it inside or outside of a higher-order effect. Every single effect library (besides the still-unreleased eff) gets certain combinations of effects + nondeterminism wrong. You could make the argument that most people don’t use nondeterministic monads, but eDSLs really shine when you have access to them, as you can turn a concrete interpreter to an abstract one fairly easily.

Looking for a few projects that make use of it, I found accelerate, hawk, polysemy and pretty-simple, so I'll be interested to poke around in their code and see how they have things set up.

Right, I think I'll just give it a shot to see. Polysemy is nice but I'm still having trouble getting what I want out of it (which may very well be entirely a fault of my own understanding)

Languages such as Koka only support algebraic effects, not scoping operations such as catch and listen. The Effect Handlers in Scope paper introduces scoping operations, which lead to the Haskell libraries fused-effects and polysemy, but they turned out to have some weird semantics. eff is her effort to fix that.

Looking at the interpreter for Reader might give some clues if this doesn't work. https://github.com/polysemy-research/polysemy/blob/master/src/Polysemy/Reader.hs#L38-L45

*But I'm not terribly well versed in Scala's other DI offerings. I came from Haskell and didn't find anything in Scala that clicked with me until I found ZIO. It reminded me a lot of my favorite way of writing Haskell programs (https://github.com/polysemy-research/polysemy)—albeit with a completely different implementation.