I recently started to fork webkit to make it a Webview with a reduced attack surface. [2]
The most interesting part was the Quirks.cpp file [1] that contains literally randomized css classnames inside the web browser code because a major website was so shitty and spec violating in their implementations.
I mean, fixing a website in a browser source code...this shit got out of hand real quick yo.
The problem with all those drafts is that Google keeps doing their own thing, and others are forced to try to catch-up or implement the same bugs/quirks that chromium does. Everything is rushed so QUICly that even Microsoft gave up at some point. And at some point in the past google realized that they can own the web if they own the Browser. And that's what they effectively do now, because the competition isn't really a competition at all anymore.
I thought you were kidding about Quirks.cpp, but it is indeed true. I can imagine a few nightmarish scenarios where a quirk that was originally a fix becomes a bug when a company decides to alter its website. Worse still, the new code will probably work perfectly in a test environment, and then fail when you deploy to your actual domain. Some developer somewhere will be tearing their hair out trying to understand what the heck is going on...
It gets worse if you search for "microsoft.com" [1] or "bbc.co.uk" in the codebase. They literally grant cross domain storage access because Microsoft's login workflow is broken in regards to CORS headers.
I mean, imagine a developer trying to fix their code behaving completely different on iOS and they cannot have a single clue why that is the case.
Additionally, WebKit is released very late to the public (sometimes even after fixes have been rolled out to iOS)...and before that nobody from the outside can even know what's going on.
From a maintainability point of view things like this should be - in worst case - a Web Extension, but definitely not be inside a Browser source codebase.
Youtube is even worse once you dig through the Apple internal plugin replacement, which effectively implements a native C++ decoder for youtube streams on iOS because otherwise you could not watch videos there. This was done before as a WebCore plugin but has been moved around into the PluginProcess source-code wise over the last two months (and currently blocks me from merging in changes, so I have to backport fixes after I removed all legacy plugin APIs).
Both ATI/AMD and nVidia use all kinds of tricks to outcompete each other since gamers only look at the fps count. That means driver optimization for specific games, even dropping fidelity a bit when that means more fps. Also, nVidia has teams that take over engine code to 'help' software devs making the most of their hardware which gives them an opportunity to push specific approaches that don't work as well on their competitors cards.
"Nvidia Gimpworks" - it isn't said in the video, but: while these techniques are slow on nVidia, they are even slower on AMD. Especially gratuitous tessellation.
https://youtu.be/AUO3LEjWsw0
You would think that they would have at least have made it into a configuration file, instead of literally hard coding rules for specific website domains directly in the browser rendering engine C++ code?
But why would they make it a config file, when they can just add it to the code?
Config files are like code, but even more shit. You never get the full set of functionality that you get with code, the compiler doesn't check your work, code navigation can't help you, and there's no debugger.
It sounds kind of funny to say “this should be in a config, not in code, but here’s a config language that lets you code your config”.
Regardless, I don’t think Dhali was around at the time Quirks.cpp was created.
Funny story, my first exposure to ruby was using it to write ant (or was it maven) config files (EJB nightmares) because the XML config sucked pretty bad (around Rails 1.0 era). It was far more concise and easy to work with, being an actual language and English readable, but I was never sure it was a good idea in a large team environment. If Java hadn’t been so cumbersome I would have used it for config as well instead of the hot garbage XML config was (is still?). I’ve seen a bajillion attempts at config languages since then. Most not great.
A lot of times 2 variables and a `map` function is all I wanted and the config would have been pleasant. Getting type errors in my editor and being able to extend record defaults was the cherry on top.
That is odd. Yet a strict check against “hulu.com” is also included, and YouTube is checked for “.youtube.com” and “youtube.com”. Perhaps this fix was to prevent malicious behavior, so they included anyone who might be phishing or abusing the API in some way? I think someone would have to dig into the quirk itself to find out more.
Note that this is very likely due to everchanging CDN domain names. If they spawn servers on demand, they're probably as messed up as googlevideo.com (which also uses hashed subdomain names).
Microsoft took a somewhat similar approach with their “ Super Duper Secure Mode” in Edge. It disables the JIT in the V8 JavaScript engine to reduce the attack area of the browser:
// FIXME: Remove after the site is fixed, <rdar://problem/75792913>
bool Quirks::shouldHideSearchFieldResultsButton() const
if (topPrivatelyControlledDomain(m_document->topDocument().url().host().toString()).startsWith("google."))
return true;
The most interesting part was the Quirks.cpp file [1] that contains literally randomized css classnames inside the web browser code because a major website was so shitty and spec violating in their implementations.
I mean, fixing a website in a browser source code...this shit got out of hand real quick yo.
The problem with all those drafts is that Google keeps doing their own thing, and others are forced to try to catch-up or implement the same bugs/quirks that chromium does. Everything is rushed so QUICly that even Microsoft gave up at some point. And at some point in the past google realized that they can own the web if they own the Browser. And that's what they effectively do now, because the competition isn't really a competition at all anymore.
[1] https://github.com/WebKit/WebKit/blob/main/Source/WebCore/pa...
[2] https://github.com/tholian-network/retrokit