Optimize CLI commands performance

[nikitabobko]

aerospace list-workspaces is slow. Well, aerospace list-workspaces is ~100ms which by itself is fine.

(1) But 100ms is the result when AeroSpace is idle. Currently AeroSpace performs all operations in one thread. Considering that Ax calls are blocking, 100ms becomes worse when a lot of Ax calls are performed.

It can be tested with alt-w = ['workspace W', 'exec-and-forget bash -c "time ~/.bin/aerospace list-workspaces" 2> ~/log']. I have 9 windows opened, and I get 300ms in ~/log. 300ms is not that nice

(2) The problem becomes worse when users need to call aerospace list-workspaces several times (E.g. to get all workspaces and focused workspaces). list-workspaces is designed to be called several times. Maybe the design could be reconsidered, or --json flag could be provided. With JSON format, AeroSpace could dump out all the info it knows about workspaces in a structured way.

Reported at: #81

[liamaharon]

Note this gets extremely bad when CPU usage is saturated (e.g. when compiling a large Rust project), Aerospace becomes completely unresponsive.

The 'fix' I have in place for now is limiting the number of CPUs available to rustc.

[CheariX]

I am using numerous aerospace CLI commands for my daily workflows (e.g., cycling through windows of active applications, focusing particular windows, etc.). The ~100 ms CLI became somewhat annoying to me since I started to feel the delay when I have ~300 ms in my scripts.

For this purpose, I experimented with some hacky cache:

on-focus-changed = [
    'exec-and-forget aerospace list-windows --focused --format "%{window-id}" > /tmp/aerospace_cache_focused.json', 
    'exec-and-forget aerospace list-windows --all --json --format "%{window-id} %{window-title} %{app-bundle-id} %{app-name} %{app-pid} %{workspace} %{monitor-id} %{monitor-name}"  > /tmp/aerospace_cache_windows.json'
]

From there on, my scripts simply use cat $file | jq.
The impact is remarkable. My scripts now feel totally different.

I was wondering if some internal caching mechanism like that would be useful in general. For example, the focused window will not change until the "on-focus-changed" event is triggered. However, if I run it 3 times in a row, each run takes approximately 100 milliseconds.
I know, this does not work for every case, e.g. a window's title could change in the background, but it could at least improve performance for all event-based changes.
(Maybe the list-* commands could have either a --fresh or a --cached flag).