CPU usage when idle

[kxt]

version: 0.10.0

I've noticed that an idle zellij process (eg. running only a barebones shell) uses quite a lot of CPU and reduces battery life too. Having multiple panes/tabs makes this more severe.

I found that in pty.rs, each fd is being actively polled to check if new data is available from a client. This can be verified by running zellij with strace, it reports thousands of syscalls per second during steady state (ie. idle).

I've created a proof-of-concept patch that uses async-std to manage io and timeout (for batching updates to reduce render time). With this patch, activity during idle drops significantly, the only activity reported by strace is os_input_output.rs's handle_command_exit looping.

I need someone to review it and discuss whether an approach like this is suitable for zellij and if so, how to move forward (eg. ServerOsApi would need to become and async trait if an approach like this would be used).

[imsnif]

@kxt - this definitely looks like a good direction! I haven't tested this out yet, but how about you whip up a PR and we'll see how it behaves?

Key things to check: performance when dumping lots of lines to the screen (eg. cat a 100000 lines file and time it).

[kxt]

Thank you for your input! I'm going to create 2 PRs:

• one is some assorted cleanup patches that I created while working on the async code, these are probably not very controversial and can be merged quickly
• the other is the actual fix for this issue

[kxt]

I've created the first PR with the refactor patches, #521.
For the actual fix, I have some problems figuring out the fakes to investigate some failing testcases.

[imsnif]

First PR merged!
Do feel free to reach out if you want a hand with the fakes. The whole testing framework is a little iffy and we're very eager to replace it :) You can reach out here or on discord, as you like.