[coldsauce]

We have a lot of ideas on how to make this problem a lot easier. It's a big reason why extensions aren't as big as they could be. Browser extension devs having to tinker and even think about the DOM or any other website-specific quirk is a leaky abstraction.

If it's an issue, people's usual approach to solving that problem is to create some mapping of "twitter-sidebar : some XPath selector" on the backend and have their extension query whenever they want to interact with the web page.

They'll have e2e tests to check if their selectors or xpaths still work. If not, they'll figure out what changed and modify the xpath. Some automate this step.

It's tricky because A/B tests are a thing, websites might change based on geographical region, and a bunch of other stuff that leads to users seeing different things in the DOM compared to the e2e test). Logging errors to something like Sentry mitigates some of this, but the complexity is still quite large.

It stems from the fact that frontends only consider humans, not robots. You either need to make your robot super resilient to change (above approach), make your robot act more like a human, have the website consider the robot ("Connect to Wallet", etc), or use the web app's backend API if it exists.

We're experimenting with all four of these approaches this year and seeing which ones are the most valuable to people, so stay tuned for more exciting extension stuff!