I could talk for days on all the peculiar bugs resolved. Once the alpha stabilizes I have drafts to publish on several topics.
You actually nailed the major pain points. Particularly window focus and state management. I've spent months solving this problem alone.
-
1. Applications data list: Getting the list is easy! Finding out which apps in that list are "real" apps isn't. Getting icons isn't. Reliably getting information on app state isn't. Finding out why something doesn't work right is as painful as can be. Doing all this in a performant way is a nightmare.
2. Applications menu renderer: Rendering the list for the menu is easy enough: the macOS app sends this data via socket. The frontend is just web sockets and web components under the hood (https://lit.dev). The difficult part was converting app icons to PNG, which is awfully slow. So a cache-warmup stage on startup finds all apps, converts their icons to png, and caches them to the app directory for read.
3. Window state: again, by far the worst and it isn't even close. Bugs galore. The biggest issue was overriding macOS core behavior on what a window is, when it's focused, and how to communicate its events reliably to the app. Although I did include a couple private APIs to achieve this, you can get pretty far by overriding Window class types in ways that I don't think were intended (lol). There is trickery required for the app to behave correctly: and the app is deceptively simple at a glance.
-
One bug, and realization, that still makes me chuckle today.. anything can be a window in macOS.
I'm writing this on Firefox now, and if I hover over a tab and a tooltip pops up - that's a window. So a fair amount of time has gone into determining _what_ these apps are doing and why. Then coming up with rules on determining when a window is likely to be a "real" window or not.
The Accessibility Inspector app comes standard on macOS and was helpful for debugging this, but it was a pain regardless.