11:20 <discocaml> <lyhokia> How do I fix this error:

11:20 <discocaml> <lyhokia> ```

11:20 <discocaml> <lyhokia> The module Stream1 is an alias for module Dune__exe__Stream1, which is missing ocamllsp [1, 26]

11:20 <discocaml> <lyhokia> ```

11:20 <discocaml> <lyhokia> here Stream1 is a file that dune can detect. I have no trouble building with dune, but ocamllsp doesn't seem to locate the source file correctly. I have confirmed that everything is installed in a same switch here.

12:04 <discocaml> <rikamg> I sometimes get spurious errors like that from the LSP, they usually go away after I `dune clean`, reload the editor and then rebuild.

12:18 <discocaml> <lyhokia> This persist, I'm assuming I have misconfiguration but I can't find out why

18:46 <parataxis> I have a thing that has a mutex that's required to be held to do certain operations. I have an unwind-protect with_lock function that calls the function with the lock held, but what I'd really like is a requires_lock decorator function that only acquires the lock if with_lock isn't on the stack. is the solution here to do some magic with domain-local storage or is there just a reentrant lock I can use?

18:47 <parataxis> or do you not even use Mutex if you're using ocaml 5 domains?

18:48 <parataxis> I'm porting C code that uses pthreads so I'm in a very threads and mutexes frame of mind where maybe I shouldn't be

19:26 <companion_cube> I use a `with_lock` but haven't used reentrant locks

19:27 <companion_cube> they're not in the stdlib, and it seems that most people interested in fancy concurrent code right now are focused on lock-free data structures

19:32 <parataxis> in this case the lock is guarding on-disk data is lock-free isn't really possible

19:32 <parataxis> think database transaction commit

19:32 <companion_cube> oh yeah I'm not saying it's the one way

19:33 <companion_cube> although for on-disk data I'd just use sqlite

19:34 <parataxis> sqlite is great but it's not the right tool for this specific thing

19:34 <companion_cube> (but do you need a file lock??)

19:34 <parataxis> this is a write-optimized search postings list

19:36 <parataxis> the way the data is laid out on disk is as an append-only log of tree roots. the only things that need to be synchronized are getting the most recent root and appending to the log but neither of those operations are atomic

19:37 <companion_cube> your quickest option might be to write a few C bindings to the relevant pthread functions (reentrant locks)

19:37 <companion_cube> and use a `with_reentrant_lock` function, I guess

19:37 <parataxis> that's an option

19:37 <parataxis> I don't know how that would interact with domains, though

19:38 <companion_cube> should be just fine!

19:38 <companion_cube> domains are just like a kind of threads (with a few more contraints on how to use them)

19:39 <parataxis> they seem like java executors

19:39 <companion_cube> I mean, if you hold locks for dozens of seconds it's certainly going to affect the GC of other domains

19:39 <companion_cube> no, they really are threads

19:39 <companion_cube> (with a slightly complicated relationship to the GC)