05:39 <d_bot> <gar> Hi folks. I've got Bazel code that supports generalized namespace libraries ("wrapped" libraries in Dune-speak). It decouples renaming from aliasing, and supports arbitrary mapping of module paths to filesystem objects. For example, here's one way you could map `A.B.C`: A: x.ml -> a.ml; B: y.ml -> foo/bar.ml; C: z.ml -> baz/buz.ml. Of course that's a pathological example, you can easily do more sensible things. You can also use the

12:52 <rwmjones> octachron: hey, any general ETA on ocaml 4.12.0 ? it's ok if not

12:58 <octachron> in the upcoming weeks: this one or the next one.

13:17 <rwmjones> octachron++ thanks :-)

15:03 <olle> new features?

15:04 <octachron> the main feature is the convergence with the multicore runtime.

15:04 <companion_cube> is the runtime closer to being reentrant, too?

15:04 <companion_cube> old dream of mine to have several OCaml cohabitate in one process

15:05 <octachron> There is mostly one item left (the major heap allocator) on the roadmap for "multicore is only additions to vanilla"

15:07 <olle> kk

15:07 <olle> what about the feature to allocate on the stack?

15:07 <olle> unboxed

15:09 <octachron> There are some propositions for more unboxed representations but there are still propositions. In term of other features, there are also names for warnings, more ! for types.

15:10 <companion_cube> "more !" ?

15:10 <d_bot> <Anurag> I'm also looking forward to the fixes done in 4.12.0 to get it working with clang11 🙂

15:10 <octachron> A quite serious performance regression fix for people tuning their GC

15:10 <octachron> !: Injectivity annotations for abstract types

15:10 <companion_cube> ohhh interesting

15:11 <companion_cube> like `type !a t` ?

15:11 <octachron> Yes: https://github.com/ocaml/ocaml/blob/trunk/stdlib/map.mli#L70

15:12 <olle> octachron: what does that mean, injectivity?

15:12 <octachron> Anurag, note that those fix should be part of 4.11.2 too.

15:13 <octachron> olle: imagine that you have 'a t = 'b t, does that means that 'a = 'b or not? The answer is yes, only if _ t is injective in its first parameter.

15:15 <octachron> In practice, this mostly matters when using GADTs where it is useful to know more inequalities

15:15 <d_bot> <Anurag> octachron: Good to know!

15:25 <Fardale> octachron: do you have an example where ! is useful with GADTs ?

15:41 <olle> octachron: ah ok

15:51 <sadiq> octachron, there's a couple more things left. I think we might even have a list somewhere on github.

15:51 <sadiq> iirc we need to do some work integrating the memory model.

15:51 <sadiq> also safepoints.

15:54 <hannes> my personal highlight in 4.12 is for sure https://github.com/ocaml/ocaml/pull/10025 "track caml_alloc_custom_mem allocators with statmemprof" :D

15:54 <octachron> Fardale, https://gist.github.com/Octachron/0ebff22f6fb006061132d6b5254e3352 : this compiles fine only in 4.12 .

16:33 <Fardale> octachron: your example did not work because M.t is abstract? Is ! similar to + in the sense that it is useful to give information if the type is abstract?

16:45 <octachron> Yes, any nominal types are automatically injectives in their parameters. The injectivity annotations makes it possible to remember this information in the case of abstract types.

16:45 <octachron> The quintessential non-injective types is "type 'a t = unit"

16:46 <companion_cube> can injectivity be combined with variance?

16:47 <octachron> Yes.

16:48 <companion_cube> `type !+'a t` ? :)

16:49 <octachron> or +!'a t. Both are allowed.

16:49 <companion_cube> so injectivity is per parameter?

16:53 <octachron> Yes, like variance.

16:54 <companion_cube> interesting, thank you!

20:58 <d_bot> <Robin> hey when i'm on a line correcting my code letters are replacing the ones which follows instead of just adding on the line do you know how to fix that?

20:59 <theblatte> @Robin: hit the "insert" (Ins?) key on your keyboard?

21:08 <d_bot> <Robin> it doesn't work bc to even when i write "insert" or anything else the text after is replaced 😦