CppCon 2019: Chandler Carruth “There Are No Zero-cost Abstractions”

The other time that you saw it was also probably me. It's from this talk, which is about how a large amount of generated protocol buffer code at Google led to a quadratic increase in compile times: [this video] . TL;DW: The reasoning is because if you use a distributed build system, then your compile time is gated by the file with the longest compile time (tail latency). The more files you have, the greater chance that one of them takes a long time. When you generate source files, you tend to produce more files than if you didn't. Most users don't use a distributed build system to compile the kernel, so on further thought, in that case compile times probably scale closer to linear with the number of translation units. But wasted cycles are still wasted cycles, and regardless of how exactly compile times scale, you should still consider the cost of longer compile times when you duplicate code. With regards to link time optimization: sophisticated analyses take superlinear complexity: [link] . Disclaimer: I work at Google.

When it comes to the overhead issues of unique_ptr, I always think of this video here: [this video] At 26:50 he gets into how this relates to ABI, and how 'fixing' some of this would require a pretty severe ABI change, as well as destructive move semantics, which would require a severe change to how C++ works. I frankly don't understand C++ well enough to fully judge about all of this, but everything he says seems entirely solid, and showcases how unique_ptr is in fact not zero-cost, and can in fact be a significant performance trade-off when you really need that extra bit of performance. > Rust occasionally forces my hand considerably harder (e.g. RC<RefCell<..>> vs std::shared_ptr<>)) in the direction of performance compromise. How so? Is RC<RefCell<T>> more expensive than std::shared_ptr<T>?

Not sure if this is sarcastic, but... 1) In dynamic languages, it's not possible to detect whether a function is used or not in the general case. For example, consider string accesses on objects. If the compiler is not sophisticated enough to resolve the set of possible strings at compile time (or such analysis would unacceptably increase compile times), then you can't shake out unused methods on that object. [1] 2) For languages like C and C++, the compiler cannot tree shake because it only knows about a single file at a time (translation unit, to be precise). You would have to rely on link-time optimizations to effectively tree shake, but LTO is not well-supported by all toolchains. Tree shaking also has a cost that I mentioned earlier -- it increases compile times. Both LTO and tree shaking in dynamic languages increase compile times superlinearly [citation needed] wrt. the size of your application. As other commenters have mentioned, it's better to avoid including unnecessary libraries in the first place. There are no cost free abstractions, and that applies to tree shaking as well. [this video] [1] For the pedants: yes, I know resolving the possible set of values (stricter than "all possible values for that type") for a variable is undecidable in the general case.

> So you're (primarily?) referring to "cognitive overhead" rather than more usual meaning of overhead as in "runtime overhead" I'm referring to both. The issue is that all that complexity makes it very difficult for programmers to write code that the compiler is consistently able to optimise to (or close to) zero overhead. It can be done with enough work and care but for your average C++ developer, they aren't going to be seeing zero overhead for a lot of the things that C++ generally leads us to assume are zero overhead. > Can you given an example? To copy the link that TwentyPosts provided (which is coincidentally also the first thing that came to my mind, albeit at a different timestamp) [this video] . The issues with `unique_ptr` in particular are that: - They force the compiler to generate exception code where maybe there isn't any necessary if you pass them into a function. You can elide this by going through and annotating things that you know can't except with noexcept however most people don't even care about noexcept in the first place and consider it line-noise. - Using move references to unique_ptr instead of pointers adds overhead by forcing the compiler to generate extra memory reads. - Passing by value should be zero cost but it'd require breaking basically every aspect of the C++ ABI to do so. Because of this you still get a double indirection which adds overhead. So your average C++ programmer probably won't do #1 which adds the largest overhead and if you need to

> Rust makes it possible to safely manage memory without using a garbage collector, probably one of the biggest pain points of using low-level languages like C and C++. In C++, memory management has not been a pain point for many years, and you basically don't need to do it at all if you don't want to. The standard library takes care of it well enough - with owning containers and smart pointers. > And Rust is famous for its optimizations in the style of "zero cost abstractions". No, it isn't that famous for those. The safety and no-UB constraints prevent a lot of that. By the way, C++, which is more famous for them, still struggles in some cases. For example, ABI restrictions prevent passing unique_ptr's via single registers, see: [link] (or see Chandler Carrouth's CppCon 2019 talk: "There are no zero-cost abstractions" [this video] )