[shiomiru]

I'm writing one for fun: https://sr.ht/~bptato/chawan/

"Fixed-width text to a grid" makes things easier (sometimes), but I think it still qualifies.

On the article itself; it might be better to start with more... basic things than the optimizations it talks about:

* Cascading & layout must be cached and optimized - far from trivial. Layout is one of the hardest parts to get right at all, to make it fast as well is another level... and I'm just talking caching, not even multi-threading.

* The "web platform" contains a very large amount of poorly thought out features harmful for user privacy and/or security. You can choose between convenience (opt out) and privacy (opt in), or try to balance the two as major browsers do. Both is often impossible.

* "Serialize the entire application state" sounds like the most complex thing you can come up with as a distinguishing feature. Much more low hanging fruit exists if you give up on writing a Chromium clone. e.g. a fun side project I've had is making my browser "protocol-agnostic", or adding a bunch small QOL stuff for myself. You can probably find new ideas easily once you start out.

* Older browsers are useful for inspiration too. The coolest features in mine are often just carbon copies of the same things from w3m, just a bit more polished. Reading about what historical engines did - even IE! - is often enlightening too. Sometimes it's just a smaller scale of what engines do now, and easier to implement. Other times it's a whole different approach.

* It's easy to get lost in the details - careful with perfectionism. As you start getting more components interacting, it will slowly become clear when the naive approach is wrong and how to improve it.

Overall, it takes time, but is also fun and rewarding - you will learn a lot, even without replacing Chromium.[0] That said, if you want to learn how "modern" browsers work, getting involved with a "modern" engine itself is probably much more productive.

[0]: To write a Chromium replacement, you may have to reincarnate as awesomekling ;)

[jwells89]

> Much more low hanging fruit exists if you give up on writing a Chromium clone.

One idea that’s crossed my mind (may never get around to trying it) is writing an engine that completely ignores everything that’s not modern HTML5 and CSS3. That’s still a lot but it seems like it’d cut down on the scope significantly. This engine wouldn’t be very useful as a browser but it’d probably be enough for displaying bespoke web content in an app or something like that.

[pavpanchekha]

It's not really clear what this even means. HTML5 and CSS3 aren't new versions of HTML and CSS that obsolete the prior stuff; they are extensions to what already existed.

So, for example, as far as I know, every web browser uses the HTML5 parsing algorithm for parsing HTML. This algorithm is very complicated, because it describes what parse tree to produce for any possible document. There's not, like, a separate HTML4 parsing algorithm; the HTML5 parsing algorithm replicates a lot of the complexities of pre-HTML5 parsing, but standardized.

Similarly, in CSS 2 the biggest complexities are things like floats, inline layout, margin collapsing, and stacking contexts. (My PhD was most of a standards-compliant CSS 2.1 layout engine.) There's not a CSS inline layout algorithm (the thing that puts words into lines) other than the CSS 2 one, though CSS 3 does add a lot more features.

In other words: the browser doesn't have a separate code path to handle old content. Instead, CSS 3 contains layout algorithms that extend CSS 2 but they don't replace them.[1] Similarly HTML 5. There are obsolete features you could probably dump (obsolete image formats, obsolete codecs) or rarely-used features (eval, document.write) or edge cases that probably don't matter that much (margin collapsing) or features where maybe your user base doesn't need it (writing direction? floats?) but this is really not so different from what the article talks about: a WebView/Electron replacement, where you commit to supporting a known universe of pages instead of the whole web.

[1] Granted some features like floats have become a lot less necessary, since a lot of their use cases can now be replaced by flex-box.

[cxr]

It's a semi-noob illusion that focusing only on modern standards and practices would surely result in a leaner implementation. This goes all the way back to the big push for web standards in 1998—e.g. you can find Slashdot Q&As from the early 2000s where people bring up the idea of how much smaller the browser could be without quirks mode and IE compatibility and then get corrected about how much of the code base this stuff actually takes up.

[zellyn]

The Ladybird folks frequently claim that directly implementing the modern versions of standards is a huge benefit.

[cxr]

It's a lot easier to implement the HTML5 parsing algorithm from the spec than trying to reverse engineer it yourself. That's a completely separate matter from the confused belief that "ignor[ing] everything that's not modern HTML5 and CSS3" would somehow "cut down on the scope significantly".

[shiomiru]

It's not a huge benefit because they are simpler; they are usually equivalent to the old standard. The benefit comes from newer standards being a lot more precise.

e.g. HTML4 did not specify what to do with invalid markup, which makes writing a conformant parser easier. In practice, many websites weren't valid HTML4, so you had to reverse engineer whatever the other parsers did with invalid markup.

HTML5 doesn't really have a formal grammar, it's specified as an imperative tokenizer and parser. It actually takes somewhat longer to implement than HTML4, but it doesn't suffer from compatibility issues.

OTOH there are new problems with the "modern" standards that old ones did not have:

* It's unversioned, updated pretty much daily; insert walking on water quote[0]. Random example: I added support for the (ancient) document.write API to my browser a few months ago. Recently I looked at the standard again, and it turns out my implementation is outdated, because there's a new type of object in the standard that must be special cased. Many similar cases; this particular one I don't mind, but it shows how hard it is to just stay fully compliant.

* It's gigantic and bloated, full of things nobody ever uses. If something gets into the standard, it typically won't ever get removed, and WHATWG has operated under this policy for more than a decade. So implementing it from start to end takes way too long, and the best strategy to get something useful is to "just" implement features that websites you use will need.

* Above is the WHATWG model, which applies for HTML and DOM, but not CSS. The W3C model (used in CSS) has versioning, but they broke it in a different way: there is no comprehensive "CSS 3 standard", just a bunch of modules haphazardly building on top of each other and CSS 2. Plus it's much less precise than the HTML standard, with basic parts left entirely unspecified for decades. See the table module[1], or things like this[2].

[0]: "Walking on water and developing software from a specification are easy if both are frozen." - Edward V Berard

[1]: https://drafts.csswg.org/css-tables-3/ still in "not ready for implementation" limbo after years.

[2]: https://github.com/w3c/csswg-drafts/issues/2452 - "resolved" by an unclear IRC log(?), but never specified to my knowledge. It's not an irrelevant edge case either, Wikipedia breaks if you get it wrong.

[immibis]

It worked for Wayland... sort of.

[pankajdoharey]

The underlying baked assumption is chromium as a rendering engine is better than Safari. How is chromium rendering engine better than safari ? Quality wise they both are same Safari actually consumes way less memory.

[roca]

I could be wrong but AFAIK Safari still doesn't have site isolation, so security-wise it's considerably weaker.

[roca]

I guess one of my points is that layout algorithms are not really part of the "most basic" decisions anymore. Replacing layout algorithms is actually a lot less disruptive to the engine architecture than switching to site isolation, say.

[shiomiru]

Fair; re-reading TFA, now I realize you explicitly instructed me to stop reading in the first paragraph :)

Trying to redeem myself with an on-topic question: isn't what you want more of a "refactoring of Blink" than "building a browser engine"? I would be surprised if a complete rewrite was really necessary for the features you want, since "saving state" already happens to some extent in all engines (even if it's just reloading from the cache) and I've seen reports about Gecko integrating multi-core cascade from Servo. What makes it hard to incrementally improve upon the current engines?

[roca]

Indeed you can incrementally improve existing engines, and certainly that's what you would try to do if you wanted one of those specific features. But I didn't write the post because I want those features, I wrote it in the hope that it might be helpful to people who are already planning to write a competitive browser engine from scratch.

Yes, for almost every conceivable features you'd be much better off adding it to an existing engine. Maybe you won't get buy-in from the core maintainers, so you'll have to maintain an out-of-tree branch, but that's still going to be much less work than doing your own engine.