Libcsp: a fast C concurrency library influenced by the CSP model | Hacker News Reader

Libcsp: a fast C concurrency library influenced by the CSP model

github.com

138 pointslime6624 comments

There is also http://libdill.org/ which has a different, better in my view, model and I have actually used in production.

I dont think libdill supports multiple CPUs, which is really hard to get right

Well it does not allow coroutines to migrate to different OS threads and handles to be shared across OS threads, but that does not mean it doesn't support multiple CPUs.

It's more correct to say that it doesn't come with that support out of the box, but it certainly does allow for it and nudges you towards a specific direction (no shared mutable state).

That has been my experience using the library, I have found that it really encourages, by means of making it the most expedient option to have write once data structures.

My use case is certainly not the common one though, so others may have different experiences.

You run multiple instances, one per CPU core.

there is a huge difference between running an instance per CPU and letting a smart scheduler orchestrate things for you.

Reminds me of this epic discussion between KirinDave, Ryan Dahl, and Aphyr: https://news.ycombinator.com/item?id=4306241

Parent my have a StackOverFlow CV.. :)

The side by side with Go is actually impressive:

https://libcsp.com/

I suspect it stops being so nice when you want to do something equivalent to go's `select` statement. Though that's purely speculation, and if I'm wrong it should definitely go in the examples!

Your skepticism made me look through the library more carefully. I came across some limitations in comparison to what Go provides:

1. As you suggest, there's no equivalent to the `select` mechanism. To perform the equivalent, you must busy loop and call `csp_chan_try_pop`.

2. It appears you cannot nest coroutines.

3. The mutex implementation does not provide an RWMutex.

4. `csp_yield` seems somewhat necessary for practical applications.

Normally I would not use another language as a basis for comparison to a C library, but that's precisely what the author of Libcsp appears to be aiming for.

All that said, there's a really slick implementation of a lock free RB queue backing this library, and It's a neat project. I think it would stand on its own better than as a "go style in C" library.

RB queue?

ring buffer queue [0]

[0] https://github.com/shiyanhui/libcsp/blob/48782baf0e63cbf2914...

People should check out the Plan 9 libthread if they are interested http://man.cat-v.org/plan_9/2/thread

The linked Wikipedia article[1] on CSP says

> communicating sequential processes (CSP) is a formal language for describing patterns of interaction in concurrent systems

The Libcsp project says it’s influenced by CSP. Does libcsp stick to this formal language?

1 - https://en.wikipedia.org/wiki/Communicating_sequential_proce...

Of course not. It implements much (not all!) of the available semantics of CSP, but does so embedded in the C language with function calls and macros. Though macros can do truly disturbing things in C, these are not enough to truly implement the syntax of CSP.

Does anyone have resources related to learning concurrency ?

I've had a very short introduction to pthreads in a CS class, but it was too succinct to get any practical benefit IMO. Knowing that a mutex can be used to limit access to a shared variable is easy enough to understand. How to structure software that runs concurrent tasks is still a mystery to me. Code I've written that uses threads appears to work but I'm not confident I haven't introduced a deadlock in there that's just waiting the worst possible timing to trigger. For instance, the pthreads manpage does a good job at listing the available functions. I have no idea when they should be used though.

If anyone can point me to learning resources around concurrency (books, presentations, talks, sample open source software that does good use of concurrency patterns, ... anything goes, really) I will be eternally grateful.

This book give a broad overview of the problem space:

https://pragprog.com/book/pb7con/seven-concurrency-models-in...

Also because I have it, here’s a distributed systems meta-reading-list. It’s old and thus I’m sure there are some broken links, but it’s a deep rabbit hole if you’re interested in distributed systems, which is basically concurrency on steroids.

https://gist.github.com/macintux/6227368

I found C++ Concurrency in Action to be pretty helpful. Also if you're able to run TSan it'll really shine a light on your code and will help you sleep more easily. (https://github.com/google/sanitizers/wiki/ThreadSanitizerCpp..., currently built into clang I believe)

Try learning Rust? It pretty much forces you to think about safe concurrency.

https://doc.rust-lang.org/book/ch16-00-concurrency.html

I wish I had a minimal version of something like this for microcontrollers (ARM Cortex M0+ and M4). Implementing event-driven code with interrupts is always a pain, even with neatly structured state machines.

You can use Protothreads[1] for this which work on anything that runs C, as they are just using setjmp/longjmp.

I would recommend FreeRTOS instead, as it has proper prempetive scheduling, context switching and wide debugger support. It's actually only 4 files[2] so not very heavyweight in either memory or stuff to learn, and has decent abstractions for cross thread communication (queues etc). Then you can write continuous code and use preemption, rather than have to code everything as a state machine and co-operatively schedule, which is painful sometimes.

[1] http://dunkels.com/adam/pt/

[2] https://www.freertos.org/Creating-a-new-FreeRTOS-project.htm...

Hmm...nice examples. Let's give this a whirl. Thanks!

Ha! Only on HN will a golem gray you down for wanting to give a guy's library a test. And I'm getting ready to do a PR in the next few days to help him out!

Reply on news.ycombinator.com