Peter Marreck (Oct 13 2026 at 14:59): Peter Marreck (Oct 13 2026 at 14:59):This is a thing from Elixir that I definitely miss in other languages that don't have it. Extremely useful when chewing through binary data of any kind.
(To see examples of it in Elixir, see: https://elixir.hexdocs.pm/binaries-strings-and-charlists.html )
Richard Feldman (Oct 14 2026 at 17:16):I think this already works with list patterns, right?
Richard Feldman (Oct 14 2026 at 17:17):[1, 2, 3, ..] =>
notramo (Oct 14 2026 at 17:52):Elixir and Gleam has some more pattern syntaxes, but I don't know whether it would be a good idea to add them to Roc.
Jonathan (Oct 14 2026 at 17:52):I think one difference is that Elixir allows you to pattern match on lengths too:
<<head::binary-size(2), rest::binary>> = <<0, 1, 2, 3>> extracts 0 and 1, so for instance if you wanted to chunk out some headers of known size, followed by the rest.
notramo (Oct 14 2026 at 17:54):I don't find it as readable, but it might be because I'm not used to it.
notramo (Oct 14 2026 at 17:54):Maybe a list matching pattern that matches an n length sublist of a List, not just one item?
Jonathan (Oct 14 2026 at 17:54):Yes it tripped me up a few times because it feels like quite special cased syntax
notramo (Oct 14 2026 at 17:55):Also, big vs little endian is not really obvious.
notramo (Oct 14 2026 at 17:57):Maybe a different syntax could be implemented, that is more readable and obvious.
notramo (Oct 14 2026 at 17:59):But I think, the best approach is to write lot of binary parsers in Roc, for different protocols and formats, then put all of them into an LLM and try out different pattern syntaxes by asking:
notramo (Oct 14 2026 at 18:01):It's very easy to design a bad language feature or bad syntax when only thinking about "How cool this feature is, look, this language has this feature!" and not about "How would it help actual code?"
notramo (Oct 14 2026 at 18:03):I'm working on a CBOR parser, and found no need for these matchers. Maybe because CBOR is so simple.
Jonathan (Oct 14 2026 at 18:16):Yeh, I have nothing to say about the need or utility of it. However, I think the motivating example for bitstrings is bit packing and working with nonstandard sizes, and destructuring this way is an ergonomic feature derived from that and in line with Elixir's general design.
notramo (Oct 14 2026 at 18:17):For example, a better way could be to define packed structs like Zig and pattern match that struct type, instead of matching the number of bits.
Richard Feldman (Oct 14 2026 at 20:43):Jonathan said:
I think one difference is that Elixir allows you to pattern match on lengths too:
<<head::binary-size(2), rest::binary>> = <<0, 1, 2, 3>>extracts 0 and 1, so for instance if you wanted to chunk out some headers of known size, followed by the rest.
this is doable in Roc too, either:
[_, _] => implies length must be exactly 2[a, b, ..] if bytes.len() == whatever => matches exactly Richard Feldman (Oct 14 2026 at 20:44):what I'm trying to say is that I think we already support everything Elixir supports for this use case, just with different syntax :smile:
Jonathan (Oct 14 2026 at 21:30):I'm not sure - if you wanted to match into a list and destructure out segments of size 3, 10, and 'rest', you would have to write that explicitly as 13 elements + rest then repackage each in new lists. Yes you could use sublist but then you wouldn't be doing it against a match target.
Jonathan (Oct 14 2026 at 21:34):A bitstring also represents an arbitary sequence of bits. If you want to slice out some bits from a list in Roc, I guess you would first represent it as List(U8) then have to do the work to marshal bits when cutting across byte boundaries?
Jonathan (Oct 14 2026 at 21:36):If it is distinct though, I'm not necessarily advocating for it :sweat_smile:
Richard Feldman (Oct 14 2026 at 23:12):Jonathan said:
I'm not sure - if you wanted to match into a list and destructure out segments of size 3, 10, and 'rest', you would have to write that explicitly as 13 elements + rest then repackage each in new lists. Yes you could use sublist but then you wouldn't be doing it against a match target.
fair, but in that case it doesn't sound like pattern matching would have any readability benefit compared to just calling sublist :smile:
Richard Feldman (Oct 14 2026 at 23:13):Jonathan said:
A bitstring also represents an arbitary sequence of bits. If you want to slice out some bits from a list in Roc, I guess you would first represent it as List(U8) then have to do the work to marshal bits when cutting across byte boundaries?
you can already do 0b1001_0111 and similar to match bit patterns on individual bytes
Richard Feldman (Oct 14 2026 at 23:15):that doesn't work across byte boundaries, but if someone has a use case where that's needed I'd want to know about the specific use case first haha
Jonathan (Oct 15 2026 at 00:05):Jonathan said:
If it is distinct though, I'm not necessarily advocating for it :sweat_smile:
Yeh to reiterate, just because I thought it was a distinct feature but not necessarily for Roc :smile:
Jonathan (Oct 15 2026 at 00:06):I think it is a natural feature in the context of Elixir's pattern matching culture
notramo (Oct 15 2026 at 09:59):It would be useful if somebody brought an actual pattern from an actual binary protocol, and showed how this matching syntax would improve the code readability or performance.
notramo (Oct 15 2026 at 10:00):Because I think it would not.
notramo (Oct 15 2026 at 10:03):(Not saying that the current syntax is the most terse or the simplest, but Elixir/Gleam syntax is not how it should be solved.)
Last updated: Oct 15 2026 at 17:56 UTC