[neonsunset]

Most modern async runtimes let you do that.

A (worker) thread dying isn't an issue in Rust, Go, C# and etc. Sure, each goes about error handling in a slightly different way either opt-in or opt-out or enforced (when unwrapping Error<T, E>) but other than that the advantages of Erlang/Elixir have faded over time because the industry has caught up.

p.s.: C# has not one but two re-entrancy syntax options - 'async/await' and 'IEnumerable<T>/yield return'. You can use latter to conveniently implement state machines.

[dmitriid]

> but other than that the advantages of Erlang/Elixir have faded over time because the industry has caught up.

It really hasn't. People fixate on the idea of just running some processes, and just catching some errors.

And yet, non of the languages that "solved this" can give you Erlang's supervision trees that are built on Erlang and Erlang VM's basic functionality. Well, Akka tried, and re-implemented half of Erlang in the process :)

But other advantages did fade: multi-machine configurations are solved by kubernetes. And it no longer matters that you can orchestrate multiple processes doing something when even CI/CD now looks like "download a hundred docker containers for even the smallest off tasks and execute those".

> p.s.: C# has not one but two re-entrancy syntax options - 'async/await' and 'IEnumerable<T>/yield return'.

What I meant by re-entrancy in the VM is this:

Every process in Erlang gets a certain number of reductions. Where a reduction is a function call, or a message passed. Every time a function is called or a message is passed, the reduction count for that process is reduced by one. Once it reaches zero, the process is paused, and a different process running on the same scheduler is executed. Once that reaches zero, the next one is executed etc.

Once all processes are executed the reduction counter is reset, the process is woken up and resumed.

On top of that, if a process waits for a message, it can be paused indefinitely long, and resumed only when the message it waits for arrives.

So, all functions that this process executes have to be re-entrant. And it doesn't mean just the functions written in Erlang itself. It means all functions, including the ones in the VM: I/O (disk, network), error handling, date handling, regexps... You name it.

[gamache]

Sometimes the advantage is not just the ability to do something -- I mean, by that logic, only bare-metal systems languages like Rust, C, Zig etc could have advantages over other languages.

Erlang (and by extension Elixir) has the advantage that the programmer can think at a higher level about their system. You don't have to write or configure a scheduler. You don't have to invent supervision trees. You can be sure that the concurrently-running parts of your system cannot possibly affect each other's memory footprint (though Rust gives a robust answer to this problem as well).

It doesn't make a perfect fit for every problem, but there is still a decent-sized space of problems -- I'd say "highly concurrent, but not highly parallel" -- where Erlang gives the programmer a headstart.

[valenterry]

The thing is, we mostly try to keep applications stateless, unless it's necessary to keep state for performance (think realtime onlinegames, hft, ...).

And in those cases there is simply no need for e.g. supervision trees because there is nothing to restart. You still need stuff like retrying, but this is supported by all major concurrency libraries / effect systems. In fact, Erlang has fallen behind here in terms of what the language offers (not what the BEAM offers, the BEAM is still top notch imho)

State is mostly moved to either the database or message queues or similar, which is pretty good.

[OkayPhysicist]

We try to keep applications stateless because state handling in most programming languages is pathological. Erlang presents a different solution to the underlying problem of state management, rather than the trying to solve the higher-level of "how best to support this one particular solution to the problem of state management".

Namely: The message queues are part of the language. They're built right in, for ease of use. Your caching layer is built into the language, for ease of use and faster performance.

[valenterry]

> We try to keep applications stateless because state handling in most programming languages is pathological.

It's true that handling state is hard in most programming languages. But that's neither the only nor the most important reason why people try to keep applications stateless.

[jayd16]

C#'s weakness here is that those two patterns are cooperative multitasking only. Under the hood they retain control of a thread until they yield execution. By default resource management is something that needs to be considered and the default thread pool is not an uncontested resource.

I don't use Erlang but my understanding is that while it is not exactly fully pre-emptive, there are safeguards in place to ensure process fairness without developer foresight.

[neonsunset]

C# async runtime is mixed mode, threadpool will try to optimize the threadcount so that all tasks can advance fairly-ish. This means spawning more worker threads than physical cores and relying on operating system's thread pre-emption to shuffle them for work.

That's why synchronously blocking a thread is not a complete loss of throughput. It used to be worse but starting from .NET 6, threadpool was rewritten in C# and can actively detect blocked threads and inject more to deal with the issue.

Additionally, another commenter above mistakenly called Rust "bare metal" which it is not because for async it is usually paired with tokio or async-std which (by default, configurable) spawn 1:1 worker threads per CPU physical threads and actively manage those too.

p.s.: the goal of cooperative multi-tasking is precisely to alleviate the issues that come with pre-emptive one. I think Java's project Loom approach is a mistake and made sense 10 years ago but not today, with every modern language adopting async/await semantics.

[jayd16]

Hey, I also prefer C# and async. Alternatives have yet to prove they can handle gui patterns where main threads matter.

...but the problems stated are real. I'm excited to hear that this might be fixed in .net 6 but it'll be a while before that rolls out to most deployments.

[neonsunset]

Apologies but it seems you have gotten wrong impression (or maybe I did a poor job in explaining).

It has never been a big issue in the first place because by now everyone knows not to 'Thread.Sleep(500)' or 'File.ReadAllBytes' in methods that can be executed by threadpool and use 'await Task.Delay(500)' or 'await File.ReadAllBytesAsync' instead. And even then you would run into threadpool starvation only under load and when quickly exhausting newly spawned threads. It is a relatively niche problem, not the main cornerstone of runtime design some make it out to be.

Also, .NET 6 is old news and has been released on Nov 8, 2021. It is the deployment target for many enterprise projects nowadays.

[jayd16]

"Everyone knows to do it right" is no protection at all. And honestly, I would push back on this in general because no its not well known at all. A fresh grad will not intuitively know to look for WhateverAsync API in case they exist and veterans will miss this as well.

Knowing that file IO is too heavy and has *Async counterpart methods is somewhat obvious to a veteran, but other long running methods are not so obvious. In this case you would need to profile your use case to understand that certain calculations/methods might be best farmed off to a different threadpool.

Unity still uses Mono and has a very low max thread pool size, for example. The thread pool is easily starved in the latest version of that engine and I'm sure it's more common than you think.

Relatively niche, perhaps, but a critical problem when stumbled upon none the less. Again, I like async/await but there are certainly foot guns left to remove.

[neonsunset]

Unity is special and has its own API and popular patterns, if you block the main/render thread it will explode, regardless of the language of choice, and Erlang/Elixir performance is not acceptable for Gamedev and will likely stumble upon similar issues.

Again, and I cannot stress this enough, we're discussing somewhat niche feature. You have to take into account that even the standard library still has a lot of semi-blocking code, simply due to the nature of certain system calls or networking code. From runtime standpoint, blocking or computationally heavy logic - there is no difference, it will scale the amount of threads to account for fairness automatically. It's that blocking just has extra cost due to being "better" at holding threads (you don't have to think about it). .NET 6 is just comparatively better at dealing with such scenarios but your app would work fine in PROD 9 times out of 10 with invalid code before or after that. It's a difference between running 'Task.Run(() => /* use up thread for no reason for seconds / minutes */))' in a 100s iterations loop going from terrible to very bad.

It's pointless to "fight against words". Just trust the runtime to do its thing right. That's why its baseline cost is somewhat higher than that of Golang or Rust/Tokio - you pay more upfront to get foolproof solution that has really good multi-threaded scaling.

If you don't want to believe the above, just look at average C# solutions on Github. There are no "special magic to learn", that's just how people write code new to the language or otherwise.

p.s.: This situation reminds me one of my colleagues who would always come up with an excuse for his point regardless of context. It's counter-productive and self-defeating.

[toast0]

> A (worker) thread dying isn't an issue in Rust, Go, C# and etc

If your Rust thread panics while it holds a Mutex, you've got a bit of a mess. Especially if it was halfway through updating shared mutable state. Probably similar in Go or C#, but I haven't used Go and only did cargo cult programming in C#, I didn't read any sources or see warnings about crashing in threads or async/await.

[rad_gruchalski]

> A (worker) thread dying isn't an issue in Rust, Go, C# and etc.

Go channels are nice but they don’t come close to Erlang message passing. In go you can’t just ignore if the channel is bounded or unbounded, open or closed. Writing to a closed channel with blow you up. It takes some time to learn it. Messages in Erl are easy fifo serial execution.