Aurélien Geron (Jun 09 2026 at 06:29): Aurélien Geron (Jun 09 2026 at 06:29):Hi fellow Roc'ers,
I've started updating the Exercism track for the new Roc compiler. I have a few exercises passing all the tests, it feels pretty good! :blush: Big shoutout to everyone who maintained this track over the past year, in particular @Anton and @Isaac Van Doren. :thank_you:
I'm currently working on a local branch which I'll push to github.com/exercism/roc very shortly. Once most (or all) exercises are migrated, and once everyone is onboard, we can merge this branch into main and go live.
If anyone wants to help port the exercises, please let me know so we don't work on the same exercises.
Some exercises depend on external libraries:
Most (if not all) other exercises should be fairly easy to migrate today.
Richard Feldman (Jun 09 2026 at 06:30):amazing, thank you so much for doing this!!! :star_struck:
Richard Feldman (Jun 09 2026 at 06:30):lmk if you hit any bugs in the new compiler, happy to unblock!
Aurélien Geron (Jun 09 2026 at 06:33):Thanks @Richard Feldman. So far I haven't run into anything serious. Perhaps https://github.com/roc-lang/roc/issues/9574 would be nice to fix, but I can workaround it so no rush at all.
Aurélien Geron (Jun 09 2026 at 07:27):What's the best place to find the list of methods available in the builtins?
In particular, I'm looking for:
Edit
It looks like U64.to_u8 is now U64.to_u8_wrap or U64.to_u8_try. The other functions don't seem to exist yet, but they're not too hard to implement, e.g., compare = |a, b| if a < b { LT } else if a > b { GT } else { EQ }
Luke Boswell (Jun 09 2026 at 08:27):https://github.com/roc-lang/roc/blob/main/src/build/roc/Builtin.roc
Aurélien Geron (Jun 09 2026 at 08:31):Thanks @Luke Boswell , I'm not sure how I missed this! That said, it might be a good idea to point to this page here: https://www.roc-lang.org/builtins/main/
Anton (Jun 09 2026 at 11:54):That page usually list all the functions like real docs :)
It's been broken because of multiple large changes to the codebase, I've been trying to fix it but keep hitting new issues
Isaac Van Doren (Jun 09 2026 at 12:43):Awesome, thanks for doing this Aurélien!
Richard Feldman (Jun 09 2026 at 13:00):Aurélien Geron said:
Perhaps https://github.com/roc-lang/roc/issues/9574 would be nice to fix, but I can workaround it so no rush at all.
fixed in https://github.com/roc-lang/roc/pull/9587
Richard Feldman (Jun 09 2026 at 13:09):Anton said:
It's been broken because of multiple large changes to the codebase
Me: "We're all trying to find the guy who did this."
Richard Feldman (Jun 09 2026 at 14:26):@Anton we should add ci/minici tests that generate the html and do some basic verification of them, like "should be at least ____ entries in the sidebar, each sidebar entry should have at least ____ html elements of this type" (where we fill in those blanks based on our knowledge of the actual number of builtins)
Anton (Jun 09 2026 at 14:32):Yeah I already added a test for that :)
Aurélien Geron (Jun 09 2026 at 22:39):I noticed that a few methods have not (yet?) been re-implemented in the new compiler, for example:
I suppose most of these will be under Iter rather than List. I don't mind reimplementing them for Exercism exercise solutions, but it would be good to know which ones are on the roadmap, and which ones have been dropped for good: if a function is on the roadmap, then I might provide it in the exercise stub.
Btw, what's the proper way to convert an iterator to a list? I don't see a to_list() method under Iter, or a from_iter() method under List.
Richard Feldman (Jun 09 2026 at 22:50):they should be in List too, just didn't get added yet! :smile:
Luke Boswell (Jun 09 2026 at 22:51):https://roc.zulipchat.com/#narrow/channel/397893-announcements/topic/Iterators/near/599963561
Anton (Jun 10 2026 at 13:43):Anton said:
That page usually list all the functions like real docs :)
It's been broken because of multiple large changes to the codebase, I've been trying to fix it but keep hitting new issues
Fixed :)
Anton (Jun 10 2026 at 13:43):https://roc-lang.org/builtins/main/
Tobias Steckenborn (Jun 10 2026 at 14:12):Funny bug here, typing anything into the search seems to lead to scrolling up :sweat_smile:
Anton (Jun 10 2026 at 14:13):Hmm, that does not happen in my browser. Can you share your OS and browser?
Tobias Steckenborn (Jun 10 2026 at 14:14):Mac, Chrome (Version 148.0.7778.168)
Anton (Jun 10 2026 at 14:18):Ok, I am able to reproduce and will make an issue
Anton (Jun 10 2026 at 14:20): Aurélien Geron (Jun 11 2026 at 09:51):Richard Feldman said:
lmk if you hit any bugs in the new compiler, happy to unblock!
I've run into a compiler panic while working on porting an exercism solution:
https://github.com/roc-lang/roc/issues/9614
See issue #9614 for more details. :folded_hands:
Aurélien Geron (Jun 11 2026 at 11:40):I've pushed the new branch with 6 ported exercises (accumulate, acronym, affine-cipher, all-your-base, allergies, armstrong-numbers):
https://github.com/exercism/roc/tree/migrate-to-new-roc
If you want to try migrating some exercises:
Then for each exercise (e.g., foo-bar):
Since some builtins are still missing, you'll have to implement them locally. Please mark them with a TODO so we can switch to the actual built-in when available.
I find it quite fun and it's a good way to catch compiler bugs.
Aurélien Geron (Jun 13 2026 at 07:37):10% of all exercises ported (12 / 111). Still a long way to go... :sweat_smile:
Aurélien Geron (Jun 14 2026 at 10:13):FWIW, here's the list of builtins that are needed for the exercises I've worked on so far. For most of them, I wrote a quick & dirty version (and I'll remove them when the builtins are available), but for the math functions, I'm stuck. I'd love to have pow_int, pow, sqrt, cos, sin.
List.map_try: 5 timesList.chunks_of: 3 timesList.intersperse: 3List.join_map: 3List.join: 3Num.pow_int: 3List.reverse: 2 timesList.reverse: 2Str.split_first: 2List.keep_oks: 1 timeList.sort_asc: 1List.sort_desc: 1Num.cos: 1Num.ln: 1Num.pow: 1Num.sin: 1Num.sqrt: 1U64.max_u64: 1Also, an Iter.to_list() function would be very nice (as well as Iter.to_set() and possibly Iter.to_dict()).
Luke Boswell (Jun 14 2026 at 10:15):The numerics are all implemented in https://github.com/roc-lang/roc/pull/9600 -- its taken me a while to land that PR
Aurélien Geron (Jun 14 2026 at 10:15):That's great, thanks @Luke Boswell
Luke Boswell (Jun 14 2026 at 10:16):TBH I suspect the failures are unrelated to the changes and more CI things we are ironing out
I looked closer at the numerics CI failure and it looks genuine, so trying to patch that nw
Luke Boswell (Jun 14 2026 at 10:16):Also Dict is upgraded in https://github.com/roc-lang/roc/pull/9643 -- but also been blocked by CI issues which I suspect are unrelated
Luke Boswell (Jun 14 2026 at 10:17):I was thinking of tackling List next but I see @Jared Ramirez has been doing a lot of work in that space so I was going to hold off a while
Jared Ramirez (Jun 14 2026 at 15:11):im almost done implemented range syntax desugaring to itrs! hoping to land asap
Aurélien Geron (Jun 17 2026 at 12:22):38 exercises done with all tests passing.
+4 done but the compiler crashes.
69 exercises to go (including 8 which rely on external libraries).
Aurélien Geron (Jun 18 2026 at 10:25):74 exercises done, 28 to go.
Aurélien Geron (Jun 20 2026 at 06:28):91 exercises done, 19 to go.
Aurélien Geron (Jun 20 2026 at 06:28):Getting close! However, the last ones are the longest, and they also include the 8 exercises that depend on external libraries.
Aurélien Geron (Jun 20 2026 at 06:31):You can check out the updated branch here:
https://github.com/exercism/roc/tree/migrate-to-new-roc
Aurélien Geron (Jun 20 2026 at 10:32):I'm not 100% confident about the right interface for some exercises, in particular error-handling and the data structure exercises, includingcomplex-numbers, rational-numbers, list-ops, custom-set, simple-linked-list, circular-buffer, and binary-search-tree.
If you want to check out some of these and tell me what you think, here's how things are organized: let's take rational-numbers, for example. Everything is located under the exercises/practice/rational-numbers directory:
.docs subdirectory contains the instructions for the users. Usually these are generic (i.e., the same for all programming languages), but sometimes there are some Roc-specific instructions in .docs/instructions.append.md so watch out for these.RationalNumbers.roc type module (which is just a stub).rational-numbers-test.roc and they will see the error messages, if any. This file is usually generated automatically using the .meta/template.j2 template, based on the standard list of tests for each exercise..meta/Example.roc, but the users never see it: it is only there to run tests when updating the Roc track, so it's not as important as the stub or the tests. Anton (Jun 20 2026 at 17:40):Can you make a WIP PR using your branch @Aurélien Geron? That makes it easy to review and comment on things.
Aurélien Geron (Jun 20 2026 at 20:33):Good point, I'll do that.
Aurélien Geron (Jun 21 2026 at 01:18):Anton said:
Can you make a WIP PR using your branch Aurélien Geron? That makes it easy to review and comment on things.
Here it is: https://github.com/exercism/roc/pull/198
Anton (Jun 21 2026 at 15:51):I have reviewed up to the grep exercise, tomorrow I will probably just take a look at your list from earlier: error-handlingand the data structure exercises, includingcomplex-numbers, rational-numbers, list-ops, custom-set, simple-linked-list, circular-buffer, and binary-search-tree.
It's a lot of code to review and it must have been a ton of work to upgrade all those exercises, thank you so much @Aurélien Geron :heart:
Aurélien Geron (Jun 21 2026 at 20:36):Thanks a lot @Anton! I'm going through your comments now.
Aurélien Geron (Jun 22 2026 at 01:47):We have a performance issue on some exercises, they take over two minutes to run (with the previous compiler they took a few seconds):
The alphametics exercise takes over 10 seconds, which is a bit too long too, but below exercism.org's threshold (25 seconds, IIRC).
Luke Boswell (Jun 22 2026 at 01:48):Is this with --opt=dev --opt=interpreter or --opt=speed etc? (it may be backend specific)
Luke Boswell (Jun 22 2026 at 01:49):Are you building these before running them roc build app.roc && ./app, or running directly roc app.roc? (it may be specific to our shims or shared memory)
Aurélien Geron (Jun 22 2026 at 01:51):It's just with zig build build-release.
Luke Boswell (Jun 22 2026 at 01:51):If it's slow on everything than it is probably bugs in the pipeline.
What command do you use to run the apps?
Aurélien Geron (Jun 22 2026 at 01:52):I run ./bin/verify-exercise nth-prime but under the hood it runs roc test nth-prime-test.roc
Luke Boswell (Jun 22 2026 at 01:53):My immediate hunch is that it might be something like the expects run at comptime using the interpreter (which is pretty slow currently)
Luke Boswell (Jun 22 2026 at 01:54):I don't have capacity to dig into this in much detail today. But I assume we would have a lot of room to improve on this.
Aurélien Geron (Jun 22 2026 at 01:54):Ah, interesting. I'll try running these exercises using roc rather than roc test
Luke Boswell (Jun 22 2026 at 01:56):So running directly with roc app.roc should use the dev backend by default now, and building roc build app.roc should use the llvm backend by default
Aurélien Geron (Jun 22 2026 at 01:57):Ah, ok, so the fastest would be roc build app.roc then ./app?
Luke Boswell (Jun 22 2026 at 01:58):yeah fastest run time, slowest compile time
Aurélien Geron (Jun 22 2026 at 02:09):Interesting: when I move the tests to the main! function, and I build nth-prime using roc build nth-prime-test.roc, then it takes too long time to build (at least 3 minutes, I interrupted it). But if I tweak the main! function to ensure that the function args depend on the main! arguments, so the compiler can't try to run the tests at compile time, then compilation takes under a second, and running the program also runs under a second. So at first glance it seems that the performance issue is when roc tries to run the function at compile time.
Luke Boswell (Jun 22 2026 at 02:11):Yeah that sounds about right
Aurélien Geron (Jun 22 2026 at 02:12):Should I open an issue?
Luke Boswell (Jun 22 2026 at 02:13):I think that would be helpful. I think @Richard Feldman has plans for comptime eval to be ran by the dev backends ... I'm not sure what the status of that is.
Aurélien Geron (Jun 22 2026 at 02:14):Oh wait, I just tweaked something in the NthPrime.roc type module, and now roc test runs in 9ms!
Richard Feldman (Jun 22 2026 at 02:14):whoa, what was the tweak? haha
Richard Feldman (Jun 22 2026 at 02:14):oh I bet it was an infinite loop somewhere
Richard Feldman (Jun 22 2026 at 02:14):(possibly in our compiler)
Aurélien Geron (Jun 22 2026 at 02:16):So the code was like this:
NthPrime :: {}.{
prime : U64 -> Try(U64, [NoPrime0])
prime = |number| {
if number == 0 {
Err(NoPrime0)
} else if number == 1 {
Ok(2)
} else if number == 2 {
Ok(3)
} else {
find_prime = |primes, index, number2| {
if primes.len() == number2 {
primes
} else {
next_index = index + 2
new_primes = {
if primes.any(|p| next_index % p == 0) {
primes
} else {
primes.append(next_index)
}
}
find_prime(new_primes, next_index, number2)
}
}
find_prime([2, 3, 5], 5, number)
.last()
.map_err(
|_| {
crash "Unreachable: list cannot be empty"
},
)
}
}
}
I had to use number2 because of https://github.com/roc-lang/roc/issues/9690 but since you fixed this issue (thank you!) I removed number2, and it fixes the performance issue.
Aurélien Geron (Jun 22 2026 at 02:17):I don't think it's an infinite loop, because the tests would pass; but they would take over 2 minutes.
Aurélien Geron (Jun 22 2026 at 02:18):For completeness, here's the new code that allows the tests to pass in 9ms:
NthPrime :: {}.{
prime : U64 -> Try(U64, [NoPrime0])
prime = |number| {
if number == 0 {
Err(NoPrime0)
} else if number == 1 {
Ok(2)
} else if number == 2 {
Ok(3)
} else {
find_prime = |primes, index| {
if primes.len() == number {
primes
} else {
next_index = index + 2
new_primes = {
if primes.any(|p| next_index % p == 0) {
primes
} else {
primes.append(next_index)
}
}
find_prime(new_primes, next_index)
}
}
find_prime([2, 3, 5], 5)
.last()
.map_err(
|_| {
crash "Unreachable: list cannot be empty"
},
)
}
}
}
fascinating! yeah if you can include all of that in the issue, would be helpful in figuring out what went wrong there :smiley:
Aurélien Geron (Jun 22 2026 at 02:19):Sounds good. :+1:
Luke Boswell (Jun 22 2026 at 02:19):@Richard Feldman is roc test dev backend or interpreter? (by default)
Richard Feldman (Jun 22 2026 at 02:20):should be dev backend but I forget if I switched it over yet :sweat_smile:
Richard Feldman (Jun 22 2026 at 02:20):in theory only compile-time evaluation of constants should still be using interpreter
Luke Boswell (Jun 22 2026 at 02:20):Yeah I remember making the PR to enable the different backends but unsure which it ran with.
Aurélien Geron (Jun 22 2026 at 02:22):Oops, sorry, I got mixed up with the multiple versions I tried locally. It's actually the reverse: the version WITH number2 runs fast, but the version without it runs slowly.
Aurélien Geron (Jun 22 2026 at 02:33):Oh actually... the 9ms might have been an illusion due to caching, duh. I'm investigating.
Aurélien Geron (Jun 22 2026 at 03:24):Yeah, sadly the 9ms was probably a caching thing, I couldn't reproduce it without caching.
I opened https://github.com/roc-lang/roc/issues/9747
Aurélien Geron (Jun 22 2026 at 21:25):103 exercises migrated, 1643/1644 tests pass! I opened an issue for that last failing test.
The 7 remaining exercises are the ones that depend on external libraries.
I'm focusing on migrating the roc-test-runner now.
Aurélien Geron (Jun 23 2026 at 09:58):I've migrated the roc-test-runner. Here's the draft PR for review:
https://github.com/exercism/roc-test-runner/pull/42
Anton (Jun 23 2026 at 12:00):Thanks @Aurélien Geron, I will check out the PR
Aurélien Geron (Jul 03 2026 at 10:40):Since the roc-parser library was migrated, I could update the sgf-parsing exercise: all tests pass. :grinning:
Aurélien Geron (Jul 03 2026 at 10:43):I also updated the code to use the new sqrt, cos, sin, e, pi, and compare functions, thanks @Luke Boswell .
Still missing a few builtins:
List.chunks_ofList.intersperseList.join_mapList.joinList.keep_oksList.map_tryList.sort_ascList.sort_descNum.is_approx_eqNum.logNum.pow_intNum.roundNum.sub_saturatedStr.split_firstAlso, perhaps add:
exp = |x| F64.e.pow(x)
Using the draft PR #31, I managed to migrate the three exercises that depend on the unicode library: anagram, micro-blog, and reverse-string. All tests pass. :tada:
Aurélien Geron (Jul 09 2026 at 18:58):108 exercises migrated, only 2 to go! :tada: These two exercises depend on the isodate library, which I'm working on migrating (but I'm travelling, so it might take a couple of weeks).
Almost all the tests pass, with just two exceptions:
binary-search-tree exercise. I opened issue #10057. All the tests used to pass.rest-api exercise fails because of a comparison bug. I opened issue #10049.We're soooo close! :smiley:
Edit: I forgot to mention that thanks to Json.parse and Json.to_str, the only exercise that depended on the json library (i.e., rest-api) no longer needs it. Pretty cool. :+1:
Last updated: Jul 23 2026 at 13:15 UTC