You need to think about what the browser is actually doing when it displays a web page. You need to think about what happens when you type the URL in and what the browser is doing conceptually.

If you have any experience programming 2D graphics with SDL this will help you to understand what is happening as well.

There is an old school talk by Nikolas Zakas on JavaScript Performance from 2009 which while some of it is out of date now due to improvements in Browser engines, the fundamental ideas are still the same.

https://www.youtube.com/watch?v=mHtdZgou0qU

If you skip to the 35 minute mark he speaks about reflow specifically. Generally I still use many of these techniques when writing vanilla JS (which is unfortunately rare these days). There are other talks where he talks about many other performance specific topics.

Obviously none of this is as good as real world experience of dealing with a thorny issue.

> If you have any experience programming 2D graphics with SDL this will help you to understand what is happening as well.

No. No it will not. Nothing in 2D graphics will help you understand things like "simple animation required full layout re-calculation and re-painting of the entire page"

Yes it will. Most of the CSS styling is literally wrappers around draw calls.

Which draw calls result in "changing border width will cause a full page re-flow in some browsers, and bot others"?

Which draw calls are "literally any animation will cause the full re-flow of the document unless you make sure an element is not a part of the document, but then you'll have issues with positioning, weird gaps etc."

Which draw calls teach you that "well, this primitive animation consumes 15% of CPU, but will only consume 6% if you go out of your way to make it 'performant'"?

You are being disingenuous. I can always use some hyper specific example to disprove any statement about what is generally true. That doesn't mean that understanding what is typically happening is useless. Which is essentially what your claim is.

> You are being disingenuous. I can always use some hyper specific example

These are not hyperspecific examples. We're literally discussing under an article which has a primitive animation which was, quote, "using 60% CPU and 25% GPU on [a] M2 MacBook"

How is 2D graphics programming helping there?

The article goes on to describe, correctly, layout properties ("The W3C spec is full of these!"), paint properties ("this tiny "bikeshed" SVG which you can find on lots of W3C spec pages. It costs ~30% CPU!") and composite properties.

These are not specific examples. These are literally footguns upon footguns that literally have no corresponding counterpart anywhere.

> That doesn't mean that understanding what is typically happening is useless. Which is essentially what your claim is.

Yes, knowing 2D graphics programming is 100% useless to understand what's happening with CSS and DOM.

> These are not hyperspecific examples. We're literally discussing under an article which has a primitive animation which was, quote, "using 60% CPU and 25% GPU on [a] M2 MacBook".

No we are not. Someone asked about how one learns some techniques that the OP mentioned (that were simpler than the ones given in the article). I gave some general advice on how one should think about what is happening and some general advice about how to think about performance.

Then you are pretending that none of this is relevant because you are concentrating on the hyper specific stuff in the article (which we are no longer talking about).

I also disagree that this cannot be reasoned about by learning a bit about how graphics works, Cartesian coordinate system and literally turning on the paint flashing tool that is in dev tools. It is literally obvious what is happening when you see it.

> Yes, knowing 2D graphics programming is 100% useless to understand what's happening with CSS and DOM.

In fact it was very helpful to help me to understand what was happening. I literally said to myself "wait a minute, that is kinda like this thing I did in SDL 1.2 in university". You are literally telling me, that something I know to be useful isn't, because you say so. That is unreasonable.

> No we are not.

Yes. Yes we are literally discussing this under an article whose introduction literally says what I quoted.

> Someone asked about how one learns some techniques that the OP mentioned

Yes, and the techniques OP mentioned literally cannot be learned from 2D graphics because nowhere in 2D graphics are you going "ah yes, to fix 60% CPU utilization for a primitive animation, you need to apply this awkward workaround that will tell the browser not to reflow the page".

Edit: and this particular technique is only applicable to this specific property for reasons that do not exist anywhere else, except in the browsers.

> wait a minute, that is kinda like this thing I did in SDL 1.2 in university

Well, it's not. Just because it kinda somewhat looks like something different doesn't make it so. Hence my question in this comment: https://news.ycombinator.com/item?id=44658482

Because those are not hyper specific, no matter how much you pretend they are

Usually you try to build something, realize it’s slow, then do a combination of searching for possible solutions, trying them, and profiling again and again to gain this knowledge.

Experience is the best teacher.

not really. the web is 30+ years of fixes, shortcuts, compatibility hacks, tech industry politics, and other assorted mistakes.

the "first principles" to understand it all is the complete set of mailing list archives of the WhatWG, plus the archives of the bug trackers of all the major browsers.

I don't think it would be different for any other system. Trying to make a piece of computer software that is completely responsive to big displays, small displays, touch screens, mice, etc is just a ton of work and a ton of small details you can't easily learn until you need to learn them.

yes. i don't mean to unfairly criticize the web as a platform, i just know it's warts because it's the platform i work on.

my criticism is reserved for the concept of "first principles", as if there should be a cohesive and simple set of rules that you can build up all the knowledge you need from. that's just not really how any complicated system works: building a comprehensive understanding of a complicated system takes a lot of work and a lot of time.

It's not "completely responsive". And literally every other GUI is just as responsive when coded for those screens.

Have you never resized a window on a desktop OS?

Not sure, but if CSS first principle is parent sets width going down to children, but sets height going up from children, then you could see how children changing its height leads to layout recalc all the way up.

So the you need to fix that height /remove resize, then google on which technique to use?

Understanding the document's flow layout, block and inline elements and build on top of that.