01:18 <discocaml> <softwaresirppi> I get that... But can't we like make an interface and swap the implementation cleverly? The user can mistakenly think he is writing brute force. That's the fantasy?

01:19 <discocaml> <softwaresirppi> Hey guys, As there are Ocaml 99 problems, is there anything else that's comprehensive and useful to solve?

01:20 <discocaml> <softwaresirppi> I'm not just asking about Ocaml, just looking for curated problem lists

01:20 <discocaml> <Kali> advent of code is fun

02:49 <discocaml> <leviroth> I don't really disagree that doing magic optimizations is good _if you can get it for free_. It's true that it makes it harder to reason about performance, but if you can still reason about _lower bounds_ of performance it's not clear why you should be sad if your performance sometimes unexpectedly exceeds those bounds.

02:49 <discocaml> <leviroth>

02:49 <discocaml> <leviroth> However, I imagine in practice it's hard to get these wins without considerable tradeoffs in complexity. For example, if OCaml silently turned `char list` into an (immutable) array, what happens when you pass the array into `List.map`? `List.map` takes a list, not an array, and the compiled form of that function assumes that the value it receives has the memory structure of a list. So now if you want to be able to support this transformation

03:09 <dh`> irc is _supposed_ to be ephemeral

03:09 <dh`> public channel logging is against the spirit

03:32 <discocaml> <softwaresirppi> Trade-offs are a very good point. Even the C compiler compiles `x % 2` to a bitwise and. Maybe we should only optimize if it's not going to affect any other behaviour

03:54 <discocaml> <yawaramin> i suppose the CCBot is logging all the IRC messages into Discord history

04:24 <dh`> oh, it is

04:24 <dh`> the bridge wouldn't work otherwise

04:25 <dh`> anyway, as far as freezing lists into arrays, it wouldn't be very hard to compile two versions of List.map and select the proper one at each call site

04:26 <dh`> the problem is knowing which things you need two copies of, and to some extent the bloat generated by unused ones

04:27 <dh`> or alternatively you could pass a sum and unpack it in the callee, but that adds other overhead

04:28 <dh`> and to really sort out when to freeze lists requires a level of intermodule analysis that's difficult to actually do in practice

06:21 <discocaml> <gooby_clown> dh`: I don't know about its inception, but nowadays I feel like IRC is used like a communication protocol/chat application like any other, and a chat history is an integral part of that. If I join the channel in the middle of a conversation here or somewhere else, I have no idea what is being discussed because I have no history. Even if you really *don't* want history, it would probably be nice to have temporary history, maybe which is del

07:35 <discocaml> <lukstafi> Would anyone be interested in the following syntax implemented via PPX? `Module_type with types instantiated` (modulo whatever the attribute or extension point is) translates as `Module_type with type t1 = t1 and type ... and type tn = tn`. The types `t1`, ..., `tn` are all the abstract types in `Module_type`.

07:37 <discocaml> <lukstafi> Ah it's not possible to do this via PPX, at least if `Module_type` is in another file...

07:57 <discocaml> <lukstafi> Is avoiding code duplication worth it if it increases total lines of code? Writing signatures and functor applications is fun but I have creeping doubts...

08:04 <discocaml> <lukstafi> Serendipitously, there's https://discuss.ocaml.org/t/icfp-and-improvement-of-module-language/15497

11:15 <discocaml> <gooby_clown> Like with many things I think it depends on the situation, it's worth it if it reduces room for error, but if it's a question of comfort only, it might be better to just duplicate

12:58 <contificate> I've got a bit of a niche question. If I was attaching to an OCaml 4 process, is there a simple set of functions I could hook to record pointer promotions?

13:01 <discocaml> <sim642> Pointer promotions?

13:02 <contificate> During a minor collection, the relocated objects will now point to locations in the major heap.

13:03 <discocaml> <sim642> The `Gc.Memprof` API might be the closest thing

13:04 <contificate> Yeah, I've seen this - I'll probably need to see how I could orchestrate it from the outside. It was indeed have a promotion callback.

13:04 <discocaml> <sim642> Looking at the OCaml implementation you can probably find the runtime C functions that it calls to set it up

13:05 <discocaml> <Alistair> I'm dusting off an old OCaml project, and when running the tests I get the following issue:

13:05 <discocaml> <Alistair> ```

13:05 <discocaml> <Alistair> File "test/typing/dune", line 4, characters 1-15:

13:05 <discocaml> <Alistair> 4 | (inline_tests)

13:05 <discocaml> <Alistair> ^^^^^^^^^^^^^^

13:05 <discocaml> <Alistair> Uncaught exception:

13:05 <discocaml> <Alistair>

13:05 <discocaml> <Alistair> (Sys_error

13:05 <discocaml> <Alistair> "/Users/ajob410/projects/dromedary/_build/default/test/typing/test_whitington.ml: No such file or directory")

13:05 <discocaml> <Alistair> ```

13:05 <discocaml> <Alistair> Looks like `inline_tests` no-longer plays nice with Dune's `include_subdirs unqualified`. Has anyone else encountered this?

13:05 <contificate> I think caml_oldify_one could be a decent hook point as well, actually - now that I read roots.c

20:54 <neiluj_> Hi! Would it be possible to restrict users outside of a library to pass user-defined implementations of a module type M to a functor taking an M?

20:55 <discocaml> <._null._> You have a functor, a bunch of given implementations and you don't want users to use any other ?

20:55 <neiluj_> yes!

20:55 <discocaml> <._null._> implementations for its input*

20:58 <discocaml> <._null._> Abstract module types ?

20:59 <companion_cube> neiluj_: hack, but you could require the functor argument to contain a value of a type that you don't expose

21:00 <companion_cube> If the module type has to be public, otherwise like._null_. said

21:00 <neiluj_> null: thanks! problem is that would hide the actual functions in the module

21:01 <discocaml> <._null._> How do you do that ? Since the type has to appear in the module type, unless I'm missing something ?

21:02 <discocaml> <._null._> Would `id` on modules be inlined to a noop ?

21:10 <discocaml> <._null._> I'm not sure, but it seems like it becomes a noop https://godbolt.org/z/h3Ef5nzKK

21:11 <discocaml> <._null._> Time to feature request private module types

21:25 <neiluj_> companion_cube: good point, and users can use the functor through a function with first-class module arg

22:43 <discocaml> <mbacarella> anyone find that they write less ocaml ever since chatgpt came out

22:44 <discocaml> <mbacarella> it really reduces the pain of using lesser/stupid languages

22:45 <discocaml> <mbacarella> sure python yeah you got me i messed up the order of these function arguments with obviously different types. good one. gpt4o please fix this for me

22:45 <discocaml> <mbacarella> dear rack of gpus, please make this python code work. thanks

22:54 <discocaml> <._null._> When you believe that all code generated by LLMs is either wrong or already available through a Google search, you don't use them for that

22:56 <discocaml> <mbacarella> yeah it's available through a google search but the LLM can copy/paste it in and change the interface to something i expect

22:58 <discocaml> <mbacarella> i'm never sure what the delta in experience is. a lot of people find them frustrating because they're wrong so much, whereas i find them useful because they're right sometimes

23:00 <discocaml> <._null._> How do you know they're right and not just more subtly wrong ?

23:01 <discocaml> <mbacarella> i kind of paste it into my editor and run it and the web form looks fine and i click submit and the data appears in the database

23:01 <discocaml> <mbacarella> looks good ship it

23:21 <discocaml> <mbacarella> like the current thing I'm doing

23:23 <discocaml> <mbacarella> I have a dashboard where users edit a template for something. there's a preview that applies a bunch of javascript transforms to the template to show what the template looks like with fake user data

23:23 <discocaml> <mbacarella>

23:23 <discocaml> <mbacarella> but when you go ahead and publish it runs a completely different python-based transform that attempts to do the same thing, server-side. obviously you don't need two of them, so I'm asking gpt4o to modify the current page to use this new /dashboard/template/preview endpoint I provided

23:23 <discocaml> <mbacarella>

23:23 <discocaml> <mbacarella> then I paste the whole web page as it exists now and ask it to fix it

23:25 <discocaml> <mbacarella> and it... worked. first try.

23:26 <discocaml> <mbacarella> it deleted all of the javascript, changed the "onchange" handler to submit to that endpoint and then populate the result in the iframe

23:26 <discocaml> <mbacarella> the only "bug" is that it provided a csrf_token unnecessarily

23:27 <discocaml> <mbacarella> /dashboard/template/preview doesn't change any data so there's no need for a csrf_token. though it's arguably good that it noticed the page had a csrf_token in play and thought to provide it

23:28 <discocaml> <mbacarella> /dashboard/template/preview doesn't change any data so there's no need for a csrf_token. though it's arguably good that it noticed the page had a csrf_token in play and thought to send it

23:45 <discocaml> <mbacarella> someone could argue with me and say sure but how hard is it to step line by line through an .html file and delete code you don't need anymore and then shove in this blob in between the script tags?

23:45 <discocaml> <mbacarella> https://pastebin.com/aG7sZhua

23:46 <discocaml> <mbacarella> like fair, that's not the hardest thing but it got a lot of details right that i don't know off the top of my head

23:57 <discocaml> <darrenldl> it only takes getting one detail wrong to throw everything off the rail