Currently on the verge of founding a (possibly viable) startup with it, but the browser itself is totally alpha for now.
Been working on parsers and protocols for a while now, and had to switch to TDD to keep my sanity together. Needed to write my own test runner that can simulate network behaviours (2G slow fragmentation is real) and peer to peer scenarios. Most servers out there don't comply with specifications, so making my own client- or peer-side implementations work was a hard task.
Currently writing my own SGML parser and optimizer, so that the browser receives only "linted and upgraded" html that is free of malicious parts, whilst embracing the idea of disallowing everything that could be potentially misused, including CDNs that do cache busting all the time.
The idea behind the browser concept is that trust is not established by default, and users should decide what website to trust, and match that with what kind of content they'd expect the website to deliver.
Anyway, IMHO, you should really focus on code clarity and hitting the high points with a good modular system. Ignore all the edge cases and if/when you open source it, that will allow people to focus on narrow pieces and make them more compliant.
The world doesn't need another rats nest like firefox and chromium have become. AKA you need to reinvent the konqueror of 1999 that spawned webkit/chromium.
I have no chance competing with google, so I’m probably gonna reuse as much of the servo project as possible when it comes to runtime and layouting/rendering. Currently a bit unplanned, on Android and iOS I have an experimental prototype up and running that’s just bundling nodejs-mobile and using a webview to localhost.
The browser UI (pwa ui) is served on port 65432 in order to allow userspace usage (ephermal ports can also be used by anyone on Windows).
Could you share more about this vision?
> writing my own SGML parser
How did you land on SGML?
What do you think of a browser/mode that parses markdown, so we can have a "markdown web" with less complex clients?
Phew, tough question. As I went into web development when XHTML 1.1 strict was the "cool shit", I kind of valued the aspect of using the web for acquiring and distributing knowledge. Not only for me, but also for publishing or other forms of media (e.g. by offering print stylesheets), screen readers, and semantic extraction of that kind of knowledge.
(I was also working on project(s) that were using DAISY to automatically convert websites into hearable formats to be consumable by blind people.)
Somehow from then (around 2000ish) to now, everything went to shit and nobody cares about that aspect anymore. News websites are too busy displaying ads and pushing subscription dialogs in my face (before I read a single line of their article) - rather than being readable or consumable.
And I kind of disagree with that. I want to make the web an automatable tool to acquire knowledge in an easy manner. And I hope I can do that programming-free. Currently, programmers can easily build scrapers - but imagine the possibilities once any person or kid can do that with a few mouse clicks.
I know there are a lot of proprietary scrapy-based solutions out there already, but honestly I think they're crappy. They see the web as DOM and not as a statistical model that a neural network "could" learn once you have a different way of rendering/parsing/modelling things.
> How did you land on SGML?
The reason why I am currently building my HTML(5) compatible parser with SGML ideas is because nobody closes tags. The spec is very complicated (especially while having an eye on what can be abused in the XSS sense or related security issues with CORS), so currently I'm kind of looking at a lot of parsers out there and try to find my own way of making this into a statistical model, so that in future my neural net adapters can optimize old HTML code into new, clean, HTML5 code.
> What do you think of a browser/mode that parses markdown, so we can have a "markdown web" with less complex clients?
Actually this was my first idea to build this. I wanted to convert all html to markdown and back, so that it's easier and cleaner. The issue I realized is that most markup and meta information that comes with a website is lost in markdown (or commonmark), and layouting sometimes implies structure, too - due to how websites in wordpress (or any user-friendly CMS) are being built.
Code-wise you usually cannot imply meaning by only looking at HTML, sadly, that's why I switched to a "filtering proxy-like" approach, whereas the Browser UI simply receives the upgraded, clean HTML, CSS (and webfonts or other assets).
I feel that one key aspect of something like this would be the ability to annotate anything on any page you stumbled across, and to navigate between all your annotations in a cohesive manner.
I'm excited to see what you make!
> (I was also working on project(s) that were using DAISY to automatically convert websites into hearable formats to be consumable by blind people.) Somehow from then (around 2000ish) to now, everything went to shit and nobody cares about that aspect anymore.
Yes, it's tragic that you could seamlessly compose streaming audio, video & text from multiple servers using an SMIL _text file_ in early 2000s, but it's all gone now.
Yet we now have large markets of broadband-connected humans with countless hours spent in front of streaming media (including video conferences) that they cannot annotate, inspect or compose. Then people wonder why they are "exhausted" after hours of Zoom meetings via powerless blackbox client apps.
There's still a tiny bit of standards activity on sync of A/V content with web text, part of the upcoming fusion of epub & the web, aligned with Google's "Web Packaging" that will enable a fully-offline internet with signed content (can of AMP worms).
https://www.w3.org/AudioVideo/Activity https://www.w3.org/community/sync-media-pub/
> so that in future my neural net adapters can optimize old HTML code into new, clean, HTML5 code.
This is exciting work. Apple has a powerful ML/AI chip on recent iPhones, likely to be used for image processing and augmented reality annotation of live video. It would be nice to apply this silicon power to the semantic ambiguity in real-world human use of markup languages.
We need an alternate timeline fork of the security aesthetic of CSS "user" vs "publisher" stylesheets, which at least tried to formalize the inherent social/power/finance conflicts between stakeholders in the web content rendering pipeline. Of course, we've since added identity, device fingerprinting, keystroke timing and countless other minutiae to the arms race. But the fundamental need for separation of powers will never go away.
Many users have powerful silicon on their devices, but today it is rarely employed in defense of "user" stylesheet/reality parsers. The proxy architecture you are developing could be combined with fully-private "user" datastores, of the kind harvested today without consent, but instead customized by the user for their own objectives, with data always in their physical control. With local personalization and ML-powered disambiguation, the unfair playing fields could be tilted a little towards local autonomy.
... and I think that this was actually the job of web browser engineers, and they failed to do so. I kind of like where Brave is going to be honest, though I do not think that an optional approach will make a change. We've been there, a lot of times, and nothing will be changed if we don't force the industries to.
Honestly currently the only Browser that is doing the right thing when it comes to privacy policies of third party cookies is WebKit/Safari [1] [2] [3] as Apple has the leverage to enforce it via their iOS market share.
Firefox/Mozilla currently is too concerned about breaking things and Chromium is a bad privacy joke outside of Ungoogled Chromium.
> The proxy architecture you are developing could be combined with fully-private "user" datastores, of the kind harvested today without consent, but instead customized by the user for their own objectives.
Exactly ;) Can't talk about this more (for now as my startup idea has to stay under the radar until Q3 this year) but I think you've figured out what I want to do with this concept.
- [1] https://webkit.org/tracking-prevention-policy/
- [2] https://webkit.org/blog/8311/intelligent-tracking-prevention...
- [3] https://webkit.org/blog/10218/full-third-party-cookie-blocki...
You might want to check out the Gemini protocol[1].
Currently, the Browser UI is actually just a PWA pointing to the nodejs instance and is reusing whatever rendering engine is available. I want to have a clean codebase, so everything is babelfree es2018 and will only run in edgium and safari 12+ (and chrome 70 or webkitgtk or webkitqt or firefox etc).
For mobile my plan is to bundle nodejs-mobile and just use a webview there, which is based on chakra (so it is JIT free and is technically allowed on iOS). For desktop I will probably unclutter servo modules and try to have a minimal fork that doesn’t have all the web apis I don’t want or need...but I’m not sure, as I’m not yet familiar enough with the servo codebase.
One thing is sure: I can’t create a competing rendering and layouting engine, so I gotta reuse an existing one.
