I'm pretty sure the font choosing algorithm takes into account the "lang" attribute as well. lang attribute and the system language are probably used to determine the fallback order. For example, on Windows 10+ with system font stack without system-ui: [0], [1]
For Han characters:
- lang="en" + system lang en -> "Chinese" font
- lang="en" + system lang ja -> "Japanese" font
- lang="ja" + system lang en -> "Japanese" font
For Hiragana, Katakana:
- lang="en" + system lang en -> "Japanese" font
which is reasonable enough but can be surprising I think. This can be confusing when you care enough about how the Han unified characters look but can't control the lang attribute (for example in Discord messages).
Side note, if the system lang is en and on Windows 10+, Japanese Github markdown text will be shown in two fonts with different "weights" [3]. It looks fine if your system lang is Japanese. I wonder if it's possible to set/guess a custom lang attribute for user/repo/markdown document.
It’s really unfortunate that the response to this blog post was to gum up CSS forever instead of fixing the browsers to do reasonable things with `system-ui`.
This is done in order to mitigate Han Unification - I suspect on OSs that aren't ancient this is probably Fine, but if you have a huge userbase like Bootstrap you probably need to support these people
> It turns out that the value of system-ui in fact not only depends on the version of the current OS, but the language of the OS as well.
https://infinnie.github.io/blog/2017/systemui.html