Meh. Runtime assertions are a poor substitute for a type system, even if you give them a type-system-like syntax.

I think it's more nuanced than that.

For a long-running program, like a network service or whatever, yes absolutely runtime assertions are a poor substitute.

For an impure program, where which branches of code are hit or aren't hit depends on the result of network calls, filesystem state, user input, etc, then yes, a type-system will be more powerful.

However, nix is more-or-less pure. By fully evaluating the output, you can hit all of the non-dead code, and hit every runtime assertion every time.

For a completely pure program, runtime and compiletime might as well be the same, running it and caching the output is the exact same as compiling it, and if it fails to run and outputs an error, that's the same as it failing to compile.

I know the "more-or-less pure" is doing a lot of work there, and in reality there's a lot of bits that aren't really pure, but it's close enough that a runtime type system gives you a lot more than you'd get from one in a traditional language.

	▲	_kb 5 minutes ago \| parent [-]
		This is a really good take. I think the core caution is this is not type-level checks. Anything this validates still needs to be eval'd. It's not a guarantee of correctness for all inputs but does look to be a fairly light (and useful) tool to make unexpected states easier for you and others to identify.

▲

anon291 3 hours ago | parent | prev [-]

Nix already has a ton written in it and it works... Magically.

Gradual typing for a system like this is the proper thing to do. You can always write a static linter for sections you need tested and maybe eventually you'll be able to type nixpkgs as well.

Any solution to the nix typing problem NEEDS graduality. Nixpkgs is too large to make it possible to have one commit that fixes everything. Starting from scratch is not a realistic option either.

▲

lmm 2 hours ago | parent [-]

Maybe you need graduality, sure. Plenty of systems (e.g. Typescript) have added a real type system that works gradually onto an existing system. While there are compromises to that approach, it's still a lot more effective than contracts IME.

▲

anon291 38 minutes ago | parent [-]

How are well specified contracts different from type script? Typescript is just annotation. Behavior cannot change. This is annotation plus runtime examinable behavior. This approach seems conducive to static checkers.

	▲	lmm 24 minutes ago \| parent [-]
		> How are well specified contracts different from type script? Typescript has types, meaning expressions can be typechecked and this is distinct from evaluating them. Typechecking is more complete/consistent/reliable - you can catch type errors that don't show up in a codepath that gets actually executed. E.g. an empty list can be well typed, which as far as I can see you can't do with contracts.