Archive for May, 2017

C++Now 2017 – Report

Saturday, May 20th, 2017

C++Now 2017 just wrapped up in Aspen, CO. A great week of presentations and discussions once more. C++Now is a very different conference from the normal mainstream. For a start, it’s small – only 150 people. And the sort of content you find at C++Now is more niche material: challenging the status quo and trying to move the medium forward and push the boundaries of what’s possible in C++.

The keynotes

This year’s keynotes focused on what C++ can learn from other languages – Rust, Haskell, and D.

Rust: Hack Without Fear! – Niko Matsakis

The Rust keynote was really good, showcasing Rust’s ownership model that allows it to achieve memory safety and freedom from data races. Pretty cool stuff, and something that C++ is just beginning to take baby steps towards.

Competitive Advantage with D – Ali Çehreli

The D keynote was very underwhelming – I didn’t come away with a good sense of where D is better than C++ and why its features are desirable.

Haskell Taketh Away: Limiting Side Effects for Parallel Programming – Ryan Newton

The Haskell keynote was probably my favourite of the trio, and the presenter did a really good job of not going down the Haskell rabbit holes so common in such talks. He didn’t try to explain monads or anything like that. He didn’t explain how return is different. He did explain real usage of Haskell and the memory model and choices for achieving side effects, and he was obviously an experienced speaker. I know some Haskell, and I was surprised to see that he put up some code on slides that was pretty advanced for a C++ crowd of mostly Haskell novices. But I think he did a great job of keeping the talk concrete, and some of the compiler technology and directions he talked about toward the end of the talk were pretty amazing.

Selected talks

C++11’s Quiet Little Gem: <system_error> – Charley Bay

This was the first talk of the week after the opening keynote. Charley is a high-energy and entertaining speaker and this talk really opened my eyes to the goodness that is <system_error> and how to use it. Around 350 fast-moving slides! I’ll definitely be using and recommending <system_error> in future.

The Mathematical Underpinnings of Promises in C++ – David Sankel

This was the kind of talk you only find at C++Now and it was great. David engages the audience really well on esoteric topics and showed how a rigorous mathematical treatment of a subject can help ensure that an API is appropriately powerful, but also that a good API shouldn’t just be a simple reflection of the mathematical operations.

He followed up this talk with its complement, a practical talk about using the API (Promises in C++: The Universal Glue for Asynchronous Programs). I’m sold: std::future is crippled, and a library like his promise library is the right way to deal with asynchronous control flow.

The ‘Detection Idiom’: A Better Way to SFINAE – Marshall Clow

This was a repeat of Marshall’s recent ACCU talk, with the opportunity for the C++Now audience to ask plenty of questions. The detection idiom (developed by Walter Brown in various talks and papers over the last few years) is a pretty simple way to do one common kind of metaprogramming operation: detecting whether a given type supports something. I feel like this is exactly the kind of thing which is going to help mainstream C++ take advantage of metaprogramming recipes: fairly uncomplicated techniques that just make code better.

The Holy Grail: A Hash Array Mapped Trie For C++ – Phil Nash

Phil presented a followup to his Meeting C++ and ACCU talk “Functional C++ for Fun and Profit“. He showed a very cool persistent HAMT implementation. There was plenty of hard data and code to digest in this talk and like all the best talks, I walked away wanting to play with things I’d just seen. He also noted that “trie” is properly pronounced “tree” (although many people say “try” which is good for disambiguation) and that he’d be sticking to “tree” not only for that reason, but because he didn’t want to be known for both “try” and Catch!

Other thoughts

The annual week in Aspen for C++Now is the highlight of my professional year. The program this year was very strong and I’m sad that I had to miss talks and make hard decisions about what to see in any given slot. But the talks are only half of this conference: at least as important is the socialising and discussion that goes on. C++Now is deliberately scheduled with long sessions and long breaks between sessions so that attendees can chat, argue, and collaborate. At other conferences I learn a handful of new things over the course of a week. At C++Now I learn dozens of new techniques, tricks and titbits at the rate of several per hour. Just a few examples of things that happen at C++Now:

  • Library in a Week: Jeff Garland corrals whoever wants to get involved, and over the week people implement really cool things. It’s like a hackathon within a conference.
  • Having a problem with your benchmarking? Take your laptop to the hotel bar where Chandler Carruth will take a look at your code and tell you exactly how the optimizer sees it, and what you can do to make it twice as fast.
  • The student volunteer program is simply amazing and inspiring. Just look at the current crop and alumni and you see young men and women who are incredibly smart and doing amazing things with C++. Figuring out new tricks and ways to stretch the language that will become codified idiom, making the library you use in a couple of years faster, cleaner, and nicer to use.

Finally, my own talk (with Jason Turner) – entitled “constexpr ALL the things!” – was very well received. I’m heading home tomorrow, but the experiences from Aspen will go back with me, leading to improvements in my own code and that of my work colleagues, making lots of things better. And the friendships I’ve made and refreshed during this week will continue that flow of information and improvement through the year.