Can it read your code and compare it to it's own and understand what is better? Can it write a "lesson learned" prompt that would make it more likely for itself to produce equivalent or better code? It's probably "no", but I'm curious.

Fixed precision is much more well-behaved. But alas, nobody could agree on it enough for it to become standard.

Je sais pas si tu veux des réponses à tes questions ou si c'est juste pour annoncer que tu veux jeter l'éponge ? Je suis d'accord que c'est complexe, sous Linux tout le monde fait son fork parce qu'on peux. C'est très fragmenté. Mais ça fonctionne, c'est juste dur de trier les info.

I remember in 2008 they already tried that. I think only the police decided to stick to it.

And then, you turn on the oven to bake your turkey and realize it's 5pm in cloudy November and all that fancy solar power is nowhere to be found.

When wind bump into a mountain it reduces the pressure on top, making humidity peak, and create rain/snow fall, which reduce humidity after the mountain and "evaporating" the cloud. Kind of like squeezing a wet cloth.

Just put the predatory men in prison.

I'm tired of having to accommodate society for people that don't respect other people.

Ideally, we could have shared bathroom and the main worry would be to hear the same fart joke and trying not to make a mess on your pants...

*sigh*

Ah, after re-reading the original post, it was indeed more about the children-grown man "issue"...

It's just not in my mind at all, especially in the restroom, it is the least sexual area...

*exhausted by the world*

There is a lot of anxiety/body image issues in the bathroom, so I can understand some people not liking them. But we still have stalls, feel free to use those!

You can put more urinals than stalls in the same space, and people spend 80% less time using them than using a stall. They are insanely efficient (5 to 10 time throughput). In a 4 stall bathroom, replacing 2 stalls with 3 urinals makes it equivalent to 15 stalls. The sinks become the bottleneck.

All spacecraft have reaction wheels that will compensate for that. Also, when you rotate the camera you will stop it before taking the picture (else it's blurry) so the rotational momentum (or lack there off) will be restored.

Wait... Should we use git to store save files. This actually makes sense (if your save format is properly indented text).

Ah you meant from a readability point of view. Yes, you are right, it's meaning will be sensitive to it the p was a pointer or the struct as a L-value

Why? It would just be a pointer-to-pointer.

Reasonable. Although one could argue the . on a pointer-to-struct should return a pointer-to-field. Then one can write *(p.v) or (*p).v or *p.v without worrying.

They are the reason why function template cannot have partial specialization and we have to create "utility" template classes as workaround.

Looks nice! Congrats!

PS: Don't do function overloading, template are superior (albeit harder), and they don't play well with each others at all.

I assume this is just engagement bait. But just in case: Zero-day exploits are not found zero-day after the release of the software. They are exploits that are being actively exploited before the vendor/provider were aware. They have zero-day to fix it. Usually they happen way past the update.

Fun fact, the "natural integers" in quaternions include half integers.

The division and modulo are multi-cycle operations anyway. And in a lot of cases you can prove a value is non-negative.

You definitely can. It simplifies a lot of issues too. Comparison between signed (pointer difference) and unsigned (object sizes) creates so many issues. A 64 bit signed number can represent any 63 bit unsigned number, but a 64 bit unsigned number cannot represent 32 bit signed numbers.

Hum, my statement might not be true on AArch64. The sign-wrap boundary is used to separate kernel space and user space, rather than stack/heap. I still think 64 bit signed everywhere is the way to go.

Fun fact: virtual memory addresses are signed (the stack take negative values, the heap positive values, and the gap always contains the sign-wrap boundary). I would recommend to only have signed numbers. And also, like Python, have / and % round rewards negative infinity (rather than toward 0)

Yeah, the calling convention are so annoying. And that's only for the arguments you are passing. Dealing with the spilling of the live value you want to preserve is annoying...

Just a hint: one of the big original issue with C++ is the single address space for everything. It causes issues with programming on CPU+GPU and also makes dependency/alias analysis hard. You may want to have "memory" classes, and pointer types being in specific memory.

I use Claude Code in the terminal of my IDE...

Fun fact, `[` is a simlink to the `test` program. It checks for it's own name, if it is `[` then it requires an extra `]` closing argument, if it is `test` then it doesn't. Most of the shells catch it and override it though.

For font rendering you will really want multichannel sdf (to preserve the sharp corners). Also, because they are annoying to generate, a friend of mine recommended me to use Chlumsky's msdfgen

Google Chrome took the dominance from the grip of Microsoft Explorer and standardized JavaScript and CSS. That's what killed Flash.

Then they made web apps a thing, and standardized HTML5.

Microsoft did all they could to stop it.

There is the "here I think lagrangian relaxation should do the trick" phase, and then there is the "rha I forgot to add that damn include file again" phase. One of those is less artistic than the other...