I recently tried to implement effects and fibers in Lua (using Lua’s coroutines ). I stumbled upon the problem that when implementing fibers through effects with a naïve implementation, then spawning a new fiber would cause the fiber to be executed in the outer context of the scheduler rather than be…

I feel like there are two issues discussed there. One issue is exhibited by the following code: [image] How to compose effect handlers with Eio Let me provide you with a simpler example. module Echo = struct type _ Effect.t += Echo : string -> unit Effect.t let run f = Effect.Deep.try_…

Eio.Fiber.both f g runs f and g concurrently in the context of the Eio scheduler (I think?). What you want is some both_inplace f g that runs f and g in the context where both_inplace was called. I don’t think anything fundamentally prevents this, I think a simple local overriding of Yield would wo…

My approach was to use a hook in the effect handling, but overriding the scheduling effects (i.e. yield) might be a much cleaner approach. However, I think that when you add the threads to the pool, things get much more complicated (implementation wise). One issue might be that if you start fibers…

A fork/spawn like that, that remembers to execute the function in the calling context and also proceeds with normal execution, would need to capture the continuation (up to the Eio delimiter, let’s say) and use it twice. OCaml by default supports one-shot continuations, and copying continuations has…

[image] monoidoid: A fork/spawn like that, […], would need to capture the continuation (up to the Eio delimiter, let’s say) and use it twice. I don’t think that’s necessarily the case. For example, one could demand that whenever handle_my_effect returns, all spawned fibers (within that contex…

[image] jbe: For example, one could demand that whenever handle_my_effect returns, all spawned fibers (within that context) will cause an exception when they are resumed, or they perform MyEffect, or even when they didn’t terminate before handle_my_effect returned. So maybe you won’t actual…

[image] monoidoid: […] it sounds like you still would need to somehow store/remember fibers (i.e. the stack segments associated with handlers) after they’ve been already used (fully popped off the stack) by the “main” path of execution. I don’t understand why that would be necessary. In my L…

I have encountered this exact problem while experimenting with an effect-based interface for logs, which enable amongst other things to define a logger in a pure manner instead of using a global ref, change the logger for a subscope, indent or add a tag to all nested logs, … let () = Logs.with_re…

[image] mefyl: I have encountered this exact problem while experimenting with an effect-based interface for logs, […] This supports me in my suspicion that it’s not just a theoretical problem (or a misconception on my end?) but something you can run into in practice as well. Thanks for sharin…

Spawning a fiber will disable effect handlers installed after main loop

Learning

talex5 December 23, 2023, 3:02pm 2

There’s some related discussion here: How to compose effect handlers with Eio - #20 by wokalski

Topic		Replies	Views
The status of eio and effects composition Community effects , eio	18	1697	November 13, 2024
How to compose effect handlers with Eio Learning	22	2461	August 28, 2023
Leaking effect handling Learning	17	594	July 17, 2026
Eio 0.1 - effects-based direct-style IO for OCaml 5 Community multicore , effects , concurrency	95	11012	February 23, 2022
Understanding cancellation (in eio) Ecosystem multicore , effects	56	5052	March 24, 2022

Spawning a fiber will disable effect handlers installed after main loop

Related topics