[galangalalgol]

But what is functional besides haskell? Purescript? Elm I guess. Ocaml is not. It has for loops even. You can write pure functional ocaml but people don't. It mattered a lot less when it didn't have true concurrency, but now clobbering things in ocaml is quite possible.

[greener_grass]

Define functional?

Even Haskell is not functional in the strictest sense. It has unsafe IO. It can throw exceptions. Functions may not halt.

[galangalalgol]

Fair. Agda, gallina, f* maybe?

My point was that without any escape hatches or magic you can code a segfault starting in ocaml5. That may be true of haskell? It is true of rust too, though the only known way to do it isn't something that is likely to happen by accident and is tracked as a bug. In ocaml5 if you use domain, it is down to experience skill, and some luck to be sure you used atomic when necessary. I'm a bad programmer despite going on four decades of experience. I'm not even remotely methodical. If I adopt ocaml for a project I'm using 4 or adding something that fails the pipeline if it finds domain anywhere.

[debugnik]

> you can code a segfault starting in ocaml5

It shouldn't, the OCaml 5 memory model bounds the reach of data races in both space and time. [1] Thread-unsafe code won't be correct when misused, but it will stay memory safe unless you reach for an additional escape hatch (or you find an implementation bug of course).

[1]: https://ocaml.org/manual/5.4/memorymodel.html

I'm much more concerned about the amount of poorly vetted escape hatches in wide use in OCaml, mainly for bindings.

[galangalalgol]

You are right! As of 5.1.1 at least it catches the cross domain access I was using to smash things. From what I am reading it sounds like it didn't work in 5.1 I could go try it in godbolt to find out when it was fixed, but I kind of don't care. Very exciting, I like ocaml and was lamenting the changes.

[debugnik]

That makes a lot more sense: The earliest 5.x releases weren't stable at all despite the non-prerelease version numbers. I waited for longer than I wanted to before upgrading from the LTS to 5, but right now it should be ok to switch as long as the few regressions, like the GC pacing issue, don't affect you workload.

[greener_grass]

Shouldn't most application programmers in OCaml be reaching for EIO or some other well-tested abstraction?

[galangalalgol]

When there was no true concurrency, only preemption, eio was safe. Not now, you can still clobber things.

Edit: i was wrong! Since at least 5.1 it catches the cross domain access and errors gracefully.

[debugnik]

Eio didn't exist before multicore OCaml actually, it was designed for it.

[phplovesong]

FP has nothing to do with mutability. You seem to lack a basic understanding what the common FP languages are, and what FP actully is.

[galangalalgol]

Pure functional programming languages do not allow mutable state. They can simulate it with monads, and they typically have impurities to allow io sode effects, but a for loop is an inherently imperative construct. Mostly functional languages like ocaml have things like ref. Functional languages aren't just about functions being composable and having no side effects, they also require that once a value is bound to a name, that it does not change. That isn't just pedantry either, it is what allows safe concurrency by default.

[phplovesong]

Ocaml is immutable. It has ref if you need mutation, this is not the default thing you grab tho. Haskell has unsafe and IORef, that does basically the same thing. Scala, rust etc has all escape hatches.

A loop by itself is not non-fp, as i can do the exact same thing via recursion. Its just syntax.

Hell, i can write a never halting program in lambda calculus with a fixed point combinator causing "undefined behaviour".

[e12e]

StandardML (standard metalanguage), scheme?

[galangalalgol]

Doesn't scheme have a set macro? I think for purity you'd have to go to something used for proofs that doesn't actually interact with the world.