Stream: compiler development

Topic: List destructure syntax

JRI98 (Jul 09 2025 at 21:41):

Hi there. Love the progress on the new compiler :ok:
I decided to check it on some old code. After a couple of modifications to modernize it, I reached the following snippet (reduced from a larger file):

module [last]

last = |l|
    match l {
        [] -> Err(EmptyList)
        [.., e] -> Ok(e)
    }

I'm wondering if the syntax is correct but I managed to craft an example that crashes the compiler :stuck_out_tongue_closed_eyes:

[1]    89899 segmentation fault  ./zig-out/bin/roc check main.roc

Luke Boswell (Jul 09 2025 at 21:45):

Match uses => now.

Shouldn't segfault though :thinking:

JRI98 (Jul 09 2025 at 21:49):

Still segfaults with => but what I was looking for was the right syntax. Thank you

JRI98 (Jul 09 2025 at 21:49):

In fact I could have just looked through the new snapshots (which are great btw). But at least the segfault gets reported

JRI98 (Jul 09 2025 at 21:55):

Hmmm don't know if it might help with debugging, but if I invert the order of the branches it works:

module [last]

last = |l|
    match l {
        [.., e] => Ok(e)
        [] => Err(EmptyList)
    }

JRI98 (Jul 09 2025 at 21:57):

This works too:

module [last]

last = |l|
    match l {
        _ => Err(EmptyList)
        [.., e] => Ok(e)
    }

Seems to happen when an empty list is the first branch.

Luke Boswell (Jul 09 2025 at 21:58):

That's a cool bug.

Richard Feldman (Jul 09 2025 at 22:19):

thanks for reporting this! I'm working on a fix.

Richard Feldman (Jul 09 2025 at 22:45):

fix: https://github.com/roc-lang/roc/pull/7986

JRI98 (Jul 09 2025 at 23:14):

Actually it doesn't seem to be fixed :thinking:
(leaving the example here again)

module [last]

last = |l|
    match l {
        [] => Err(EmptyList)
        [.., e] => Ok(e)
    }

Luke Boswell (Jul 09 2025 at 23:33):

It looks related to nested patterns specifically

Luke Boswell (Jul 09 2025 at 23:34):

It's fine with either just a lambda or just a match, but both tother and we are re-using the same scratch for patterns (my current theory)

Brendan Hansknecht (Jul 09 2025 at 23:38):

This is why we need to clean up all the low hanging fuzzer crashes. Then it can reach more meaningful failures like this

Luke Boswell (Jul 09 2025 at 23:44):

Apparently our segfault is in the type-checker ... specifically with empty lists

Luke Boswell (Jul 10 2025 at 00:21):

Ok, I've got a fix... I'll push it up soon.

Luke Boswell (Jul 10 2025 at 01:06):

https://github.com/roc-lang/roc/pull/7988

Last updated: Jul 26 2025 at 12:14 UTC