Every single time.

It's a tired argument, and we should strive for the intellectual humility to do better in this forum.

Cargo, please PLEASE give me a way to disable third-party build.rs and whitelist the ones I need. And please loudly mark any update that adds a build.rs where there was none before.

`cargo add` is sufficient to compromise you, before you have a chance to even vet the code.

I don't have direct HTTP/1 vs HTTP/3 comparisons, because 0% of our requests are over HTTP/1 today...

The secondary problem is SPAs approximately double the amount of code you need to write compared to a server-side approach for the same level of interactivity.

Once you have sufficient context in your html, this is no longer necessary.

Our webapp is achieving 800:1 to 5000:1 compression ratios pushing down the entire HTML page over brotli and zstd. The first render is about 20:1 compression, every interaction after that is 800-5000:1.

We are using datastar with the recommended "fat morph" approach where we re-render the entire HTML page from scratch on every request. Our typical RTT is <100ms including rendering, then the brotli / zstd compression window cuts down the wire size to literally _bytes_.

This is SO much more efficient than our previous reactjs & graphql render waterfalls with state duplication in the client side. All states lives on the server, all renders are authoritative, and we need to maintain roughly half the code size for better first-render and massively better interactive performance. Appserver and DB pressure is massively reduced too.

(This is a medical device with a highly interactive UI, image viewers, editors, not a boring old static website.)

IE6 was 97%, how did that work out?

And they have failed every single time.

The bitter lesson essay was written precisely to dismiss that approach, which used to dominate conferences and scientific publications of the era. A general learning system, given sufficient computation power, will always outperform specialized crafted systems in the long run.

Think of it like this: if a world model is a useful abstraction, the general learning system will create it by itself during its training, without us needing to implement it by hand. This is the bitter lesson. And it comes for us all.

You use Tailwind in the same way you use BEM: becuase it makes it easier for frontend developers to write correct CSS.

I simply answered his question - tailwind is indeed just plain old CSS with a more ergonomic DX.

My global tailwind.css defines two sets of css vars for the light and dark theme, and it all gets applied globally with a single line in my root html template. I don’t need to touch my markup. Tailwind simply compiles this down to a single css file in the end.

That handles consistent sizing, spacing, text, colors, dark mode, and anything else your design system needs. It all compiles down to css in the end, so no runtime overhead and gives you the superior tailwind dx during development.