Sabitlenmiş Tweet
Steve Downey
22.8K posts
Steve Downey
@sdowney
Software engineer at Bloomberg LP Views are my own he/him Parody of a real software engineer and grown up. @[email protected]
New York, NY Katılım Temmuz 2007
1.1K Takip Edilen1.1K Takipçiler
Steve Downey retweetledi
🤔 What if writing faster C++ code actually meant writing *simpler* code?
Incredibly excited to share that I'll be giving my first-ever keynote at @CppCon 2025:
➡️ “More Speed & Simplicity: Practical Data-Oriented Design in C++”
cppcon.org/2025-keynote-v…
#cpp #cplusplus #gamedev
English
@__phantomderp Sure, but C++ still gets to check the recursive box.
And if C needs to spell the type _SelfFunctionPointer and it only works for non-captures, that's a C problem?
English
@sdowney C does not have templates, so that doesn't help here.
English
Alright. Finished the introduction / motivation section and have a clear path.
Let's get to work.
English
@__phantomderp Parallel, these days, is usually the same operation at the same time, where concurrency is different operations all making progress at the same time.
With "same time" being very underspecified.
Threads are a different axis and may or may not be involved with either.
English
I always forget whether it should be "these are never invoked in parallel" or "these are never invoked concurrently". I think the proper terminology for "not at the exact same wall-clock physical time" is "these are not invoked in parallel"....
English
Steve Downey retweetledi
C++26 is going to have Reflection!
I am very excited about that fact, so I wrote a new #cpp post with a very cool example. We're taking a JSON string and turning it directly into a C++ object:
brevzin.github.io/c++/2025/06/26…
English
@Nekrolm @__phantomderp Which one do you mean? It's only a pointer, and there isn't an extra bool, if that's the one.
It might be possible to be technically conforming to have a bool, but no one hates their customers that much.
English
@__phantomderp Cool, will it have niche optimization?
English
Thanks to Steve Downey and others, optional<T&> officially adopted into C++26.
It would've been great during C++17, but the uphill battle had to be fought first. But, hey. The first best time was 9 years ago,
the second best time is now.
English
@_mort96_ @__phantomderp It was never a Container, which means a bunch of requirements in the library, but we did make it a range, and a view, for C++26.
`for (auto t&& : opt_T) {...} ` works
Many people hate that syntax, though, because iteration of 0 or 1 bothers them.
English
@Nekrolm It's mandated to have that optimization, yes.
English
@Nekrolm @__phantomderp If you can use C++20 the reference implementation is at
github.com/bemanproject/o…
(There's a constexpr bug in a test case, only for 20, I will fix before marking it 1.0)
English
@__phantomderp The really important thing, and why I worked on it, is this is also the semantics for variant over references.
variant<T...>& ought to be the iterator for tuple<T...> but tuple can hold things that variant can't.
Optional is a more efficient variant<T, monostate>.
English
@__phantomderp github.com/bloomberg/clan…
I think it's up to date.
Mod whatever we've changed this week.
English
I need to get my hands on a build of reflection for C++ so I can finally evaluate it with a real code base.
English
@__phantomderp But what if I want to do it again, only slightly worse?
English
It's finally done.
I've drawn blood from the win32 stone.
I just need to write and document all of this, so nobody ever has to do it again.
English
@FleetingShadow @__phantomderp Together with reflection and consteval you can write a parser for other programming languages and compile them into C++. At least in theory.
English
@__phantomderp *applause*
This was a bigger hole than people think. Fixing it makes writing all kinds of table-based code a LOT easier.
English
Steve Downey retweetledi
Preprocessor embed is in C++26. You can go bother your vendors to put it in. The offset parameter may make it through as well, meaning it will go in fully complete and not need a second standards revision to patch the offset(3) parameter in.
It's finally, truly, over.
English
@__phantomderp @qdoit_ Can't wait for the explanations of why C++ can't.
English
@qdoit_ Speaking of that, we're getting closer to getting a guaranteed form of that in C soon, baybeeee 🥳
English
I gotta see so many posts saying "Rust is slower" and like I think these people just evaluate languages and designs based on vibes and not based on like the material output from the compilers.
English
@headinthebox OO is great because the real world is filled with objects, not values.
OO is awful because state and behavior are intractable problems.
English
I don’t understand the hate. A function is an object with just a single method. Algebraic types are sealed inheritance. FP is a subset of OO.
Dmitrii Kovanikov@ChShersh
I was working on an ANSI terminal formatting library recently and discovered a lucid example of a simple problem to demonstrate why FP is superior to OOP. I might even blog about it. OOP is cooked.
English
More seriously: defer is an improvement, but Lina is right: it's not systemic. defer is good for one-offs, but still requires you to Always Do It Correctly. You can paste this over with macros that always pair the creation with the defer setup, but... alas.
English
Hh... haha... yeah.... who would wanna put "defer" in a language like C when you can just have Drop/Destructors from Rust/C++, haha. Hahaha......
Haaaaahhhhh........
Hoshino Lina / 星乃リナ 🩵 3D Yuri Wedding 2026!!!@Lina_Hoshino
All those things Rust has that Zig doesn't are important for the things I'm doing. Destructors/RAII are fundamental to how the driver tracks and cleans up firmware structures safely and reliably when needed. If I had to write "defer" everywhere it would be a bug-prone mess...
English
@blelbach @jfbastien I've had the pleasure of programming 28xx DSPs that have a 16bit byte.
For fun.
It did use a sensible representation, though.
English
Pretty sure @jfbastien wrote a proposal suggesting we do just that, specifically for byte size.
Sebastian Aaltonen@SebAaltonen
C++ standards committee should forget about 50+ year old computers. There's indeed couple of people around the world maintaining ancient software, but those people have not updated their compiler for decades. Nobody will try to use C++20 on those systems. It's safe to drop them.
English
@hankadusikova Graph tagging in expression evaluation. Basically the same thing as Mark/sweep GC, but for reduction in expression trees, distinguishing reduced value and unreduced expression.
English