Styles injected into "author" origin layer share context with "page", so are exposed to their scripts, and when we need to beat some author's !important rule, we must make sure we have equal or greater selector specificity, what leads to absurd selectors like `*:not(#a#b#c#d) { border-radius: 0 !important }` with hope that page author will not use matching selector with five IDs.
In regular User Stylesheet any important rule wins over all author rules, regardless their selector specificity and importance. I find this design decision as very elegant and effective distribution of power.
Do you have more details? I fricking hate how extension injected CSS
"content_scripts": [{"css": [xxx
somehow lands in this weird immutable state where Devtools doesnt let you edit any of it, really slows down development.
It is true that this part of specification was never developed to the degree to be really practical, but luckily at least some implementers went ahead and added what was missing: ability to target specific styles with specific chunks. UserCSS grew from this very mechanism.
I'm sure Firefox can do it even today (with one pref flipped), I guess most "good" progressive browsers as well, but last time I checked Chrome-ium chopped it out around 2014, effectively violating few specs [41091633]. They later provided the mechanism for extensions to use (the ability to inject style into cascade on the "user" origin level) but to my knowledge no extension picked it up as of now. (Maybe uBlock, after all?)
[41091633] https://news.ycombinator.com/item?id=7457970