Stream: announcements

Topic: Breaking change: removing Iter.rev and Iter.take_last

Richard Feldman (Aug 08 2026 at 01:45):

we currently have an assumption that iterators are reversible, but I think that was a mistake in hindsight. Some custom iterators simply aren't (e.g. an infinite range), or have terrible performance if you reverse them (e.g. a linked list), and I think it's better to just expose things like List.iter_rev from only the data structures that can reasonably offer reversed iterators, instead of having the reversal operation in Iter itself.

Richard Feldman (Aug 08 2026 at 01:46):

(I haven't landed his change yet, but I'll be opening a PR for it soon.)

Aurélien Geron (Aug 13 2026 at 02:15):

This did indeed break a few exercises, I'm fixing them now.

One use-case was: (1..=10).rev() which no longer works.
It would be nice to have a simple way to express this. For now, I've replaced this with List.from_iter(1..=10).iter_rev(). It works, but it materializes a List when we don't really need one.

Aurélien Geron (Aug 13 2026 at 02:31):

A few random ideas:

• 10..>1 (but only works with exclusive ranges and step -1)
• 10..=1 by -1
• 10..-1..1 (F# style) or maybe 10..[-1]..1
• 10, 9, ..=1 (Haskell-style)

However, none of these fix the fact that these ranges should be reversible.

• So perhaps 1..=10 could produce a Range type. You would have to add iter() or rev_iter() to actually get an Iter, example: (1..=10).iter() to get 1 to 10. Might be a good compromise actually. It would also allow things like (1..=10).step(2).iter_rev() to get 10, 8, 6, 4, 2.

Also, (1..=10).size_hint() could return Known(10) (currently it returns Unknown).

Richard Feldman (Aug 13 2026 at 02:40):

Aurélien Geron said:

perhaps 1..=10 could produce a Range type. You would have to add iter() or rev_iter() to actually get an Iter, example: (1..=10).iter() to get 1 to 10. Might be a good compromise actually. It would also allow things like (1..=10).step(2).iter_rev() to get 10, 8, 6, 4, 2.

Also, (1..=10).size_hint() could return Known(10) (currently it returns Unknown).

yeah, that all seems reasonable! We could put it under Num.Range so it doesn't need a separate top-level thing

Richard Feldman (Aug 13 2026 at 02:40):

and since for works on anything that has an iter method, Range would continue to Just Work inside for loops

Aurélien Geron (Aug 13 2026 at 02:47):

Oh right, I had forgotten about that last point, cool. :+1:

Luke Boswell (Aug 13 2026 at 02:56):

@Aurélien Geron want to make a GH Issue for this?

Aurélien Geron (Aug 13 2026 at 03:09):

Sure, I'm on it

Aurélien Geron (Aug 13 2026 at 03:32):

https://github.com/roc-lang/roc/issues/10753

Aurélien Geron (Aug 13 2026 at 04:42):

Note: this change broke roc-random, so this PR fixes it.

Luke Boswell (Aug 13 2026 at 05:07):

I'll wait for tomorrow's nightly and then bump that and kickoff CI and a release

Aurélien Geron (Aug 13 2026 at 05:08):

This roc-random? https://github.com/kili-ilo/roc-random

Luke Boswell (Aug 13 2026 at 05:09):

yes, the latest nightly doesn't have the Iter thing does it?

Aurélien Geron (Aug 13 2026 at 05:09):

Ah yes, indeed

Luke Boswell (Aug 13 2026 at 05:09):

I had another PR staged to update the nightly, but then I realised it didn't have the fix in it

Aurélien Geron (Aug 13 2026 at 05:10):

Btw, something seems to be broken with types, see this new issue.

Aurélien Geron (Aug 13 2026 at 05:12):

I also ran into a type issue while fixing roc-random: I had to add a type spec for component_generator on line 278 in package/Random.roc`, or else it ran into a type issue.

Luke Boswell (Aug 13 2026 at 05:12):

this will be a good regression test to add when we fix that

Luke Boswell (Aug 13 2026 at 05:13):

Aurélien Geron said:

I also ran into a type issue while fixing roc-random: I had to add a type spec for component_generator on line 278 in package/Random.roc`, or else it ran into a type issue.

Do you think you could make a minimal repro? I guess it's probably ok to make an issue with the git commits from roc-random, but would be nice if we had something small we could convert into a regression test

Aurélien Geron (Aug 13 2026 at 05:14):

I'll give it a shot.

Last updated: Sep 03 2026 at 15:16 UTC