CppCon 2016: Nat Goodspeed “Elegant Asynchronous Code'

—
—
Asynchronous I/O is an essential technology for increasing concurrency without introducing more threads, which carry both overhead and risks. Unfortunately, leveraging async I/O can be disruptive to careful code architecture.

This talk focuses not on the mechanics of async I/O, but rather on a library that manages async I/O with code that looks and behaves like traditional sequential code. Boost.Fiber introduces "userland threads," permitting the application coder the benefits of both async I/O and clean layers of abstraction.

This talk presents an overview of the library, discusses integration with event-driven frameworks and illustrates several useful patterns.
—

Nat Goodspeed
Linden Lab
Nat Goodspeed first read about C++ shortly after implementing a medium-sized language interpreter in classic C. Almost every C++ feature led to forehead smacking accompanied by muttered remarks about how the language could have solved THIS problem... He has been fascinated (sometimes infuriated) by C++ ever since.

When he first encountered Boost libraries, a light shone from the heavens and choirs of angels sang "AAAAAAH..." The idea of writing and maintaining C++ code without Boost has become unthinkable -- even now that some of those libraries have been adopted into the C++ standard.

Nat has presented papers on coroutines and fibers to WG21, the ISO C++ committee. He has spoken on these topics at C++ Now conferences.

Nat has worked on databases, games and virtual worlds, usually on architecture and infrastructure. He has functioned as Boost evangelist within various organizations.
—

*-----*
*-----*

[XxxGuitarMadnessxxX]

5 years late to watching this video but my God has this actually helped me understand the usefulness of the newly added coroutines with async functions.

[ZiyangJiangSheep]

To me, this is a very good presentation of why we need fiber. Try to use boost asio, boost coroutine before watching this video.

[nullplan]

24:00 You can manage stacksize and set stacks entirely in pthreads as well.

[DonPhrostByte]

I still feel like vc++'s resumable functions are easier to reason about, but this did show me that perhaps fibers still have some room to innovate.

[SlowCarToChina]

I must have missed something: what is the advantage of using fibers instead of threads?

[YourCRTube]

Good talk, but skimmed though the examples a bit too fast. It is also kind of confusing to me,  that all examples are ultimately blocking to the caller, or I am missing something?

[PaweSkalczynski]

Introduction was a bit lengthy, but apart from that - nice talk ! Thank You!

[sdmarlow]

20 minutes in only to find boost lib is part of the solution.

[mwont]

I think fibers are the closest you can get to python asyncio in c++.

[dartme]

I still don't understand how to accomplish "elegant asynchronous code". Maybe I'm trying too much to fit this to my own use case that doesn't make sense? I'm thinking of a networking library that I use in which I register callbacks that are called when certain network traffic is encountered. This callback stuff is making my code look like spaghetti. Perhaps this fiber-based approach to asynchronous code isn't helpful in my case because the library wasn't written that way. It creates a new thread on a connection and calls my callbacks through that thread, so I think I'll need to go the message queue route.

I'm still happy I watched the talk: I learned a bunch about boost fibers!

[towitorstenwill]

Could you get rid of the line by line presentation, next time? That would improve the presentation style.