00:00 <discocaml> <froyo> where's the document that describes the current GC implementation? I remember reading the retrofitting paper and the ocaml-multicore notes but I think the merged & released runtime deviates slightly from those?

01:25 <discocaml> <tautologico> now that ocaml has multicore, it could be interesting to have an ocaml version of this benchmark comparing multiple languages https://pkolaczk.github.io/memory-consumption-of-async/

02:32 <discocaml> <froyo> I've done 16M tasks easily with Domainslib.Task

04:41 <discocaml> <froyo> back with numbers:

04:41 <discocaml> <froyo> baseline: 4 MB

04:41 <discocaml> <froyo> 10K: 10 MB

04:41 <discocaml> <froyo> 100K: 30 MB

04:41 <discocaml> <froyo> 1M: 250 MB

04:41 <discocaml> <froyo> Time overhead: ~4 seconds

04:42 <discocaml> <froyo> one problem I faced was nonblocking sleep

04:43 <discocaml> <froyo> ```ocaml

04:43 <discocaml> <froyo> let tasks = Array.init 100 (fun _ -> Task.async pool (fun () -> Unix.sleep 1)) in

04:43 <discocaml> <froyo> Task.run pool (fun _ -> Array.iter (Task.await pool) tasks)

04:43 <discocaml> <froyo> ```

04:43 <discocaml> <froyo> this blocks for ~34 seconds on num_domains:2

04:45 <discocaml> <froyo> as long as each task's heap allocation doesn't happen when we await on it, then omitting the sleep should be fine. Instead, to block the tasks, I used a Chan.recv for each task.

06:02 <discocaml> <darrenldl> it'll hopefully be easier once i actually finish the PR for eio task pool properly...

06:08 <discocaml> <tautologico> so better than all others except Rust+tokio?

06:46 <discocaml> <hockletock> the author of that post says he used ChatGPT to create implementations in languages he doesn't know, that's a huge grain of salt for any kind of usefulness of the benchmark

06:51 <discocaml> <tautologico> true

07:24 <discocaml> <bellamy> Maybe this is obvious, but what does the Co refer to or stand for in Cohttp? Scoured the docs and have not find the answer. (Conduit, Concurrent, Coroutine, other?)

08:01 <discocaml> <Le condor du plateau> actually it stand for comunism

09:05 <discocaml> <darrenldl> i kinda like the idea of using chatgpt to fill in code for langs you don't know, cause eventually some experts will get annoyed and give you the expert version

13:34 <engler> I have to convert a lexer/parser written in OCaml with ocamllex and ocamlyacc into a different programming language. Are there ways to convert it/interface to it?

13:41 <octachron> What do you mean by "interface to it"? It is possible to call OCaml function from C, thus you can call the OCaml parser from a different programming language.

13:47 <engler> octachron: Yes, this is what I meant with interfacing to it

13:48 <engler> Although I would prefer a different way to be honest

14:03 <octachron> You can use menhir and ocamllex parser to parse the lexer itself and try to generate the equivalent code in your lexer/parser generator of choice. Or try to write new backend to ocamllex/menir. For small/mid-sized grammar that needs to be converted once, doing the conversion by hand is likely to be faster.

14:06 <companion_cube> So it seems that the Mutex module uses regular pthread mutexes. I wonder if the fine people spending a lot of effort on lockless structures would also do a futex based implementation of Mutex, since it's not like it's going to become useless...

15:38 <discocaml> <bellamy> Could also be Coinductive? Here is a paper in the Caml world which uses the Co prefix with this meaning: https://www.cs.cornell.edu/~kozen/Papers/CoCaml.pdf

15:50 <discocaml> <Anurag> Reading the description from an older version of cohttp, `co` is most likely referencing `cooperative concurrency` (https://github.com/ocaml-attic/cohttp.old)

18:20 <discocaml> <bellamy> Nice find, thanks! Also just realized that Alcotest must stand for (A) (L)ightweight and (CO)lourful test framework per its description.

22:13 <discocaml> <Et7f3 (@me on reply)> It is also a pun to detect alcohol

23:33 <dh`> anyone know what the deal with the permissions argument of Sys.mkdir is?

23:33 <dh`> it does not appear to be numeric unix permissions

23:57 <companion_cube> You sure? Are you using the octal notation?