cargo-call-stack VS asyncly

Yes, it's what I wrote about in the last paragraph. If you can compute maximum stack size of a function, then you can avoid dynamic allocation with fibers as well. You are right that such implementations do not exist in right now, but I think it's technically possible as demonstrated by tools such as https://github.com/japaric/cargo-call-stack The main stumbling block here is FFI, historically shared libraries do not have any annotations about stack usage, so functions with bounded stack usage would not be able to use even libc.

For rust code, I have found https://github.com/japaric/cargo-call-stack to be the best available option, as it does take advantage of how Rust types are implemented in LLVM-IR to handle function pointers / dynamic dispatch a little better. An even better solution would try to use MIR type information as well to further narrow down targets of dynamic calls in a Rust-specific way, but no such tool exists that I know of.

cargo-call-stack Static stack analysis!

Generators can just dump stuff on the stack. They have additional their own stack for storing their state. If you can prove an upper amount of creation of generators in the call graph, that would however work. There is for example this nice tool for Rust doing the overapproximation.

Not easy at all.

I know that in the small-embedded world, people do work on such things.

Eg https://github.com/japaric/cargo-call-stack

One of the main benefits of async/await in Rust is that it can work in situations where you don't even have threads or dynamic memory. You can absolutely use it to write very concise code that's waiting on an interrupt on your microcontroller to have read some data coming in over I2C from some buffer. It's a higher level abstraction that allows your code to use concurrency (mostly) without having tons of interactions with the underlying runtime.

Every major piece of software that I have worked on has implemented this in one form or another (even in non-modern C++ where you don't have any coroutine concepts, Apple's grand central dispatch,). If you don't then your business logic will either be very imperformantly block on IO, have a gazillion of threads that make development/debugging a living hell, or be littered with implementation details of the underlying runtime or a combination of all 3.

If you don't use existing abstractions in the language (or through some library), you will end up building them yourselves, which is hard and probably overall inferior to widely used ones (if there are any). I have done so in the past, see https://github.com/goto-opensource/asyncly.

Keep in mind that these is a really basic building block where you can bring your own runtime and hook coroutines into it, not something that is at all usable out of the box. This is exacerbated by the fact that the C++ standard library is still lacking support for non-blocking futures/promises.

To see how it can be used for actual asynchronous operations on a thread pool, take a look at asyncly, which I co-authored:

https://github.com/LogMeIn/asyncly/blob/master/Test/Unit/fut...