Hopefully this change will bring some responsiveness to the LSP plugin and other things. I remember LSP performed better when I first developed it but at some point lite-xl introduced a change that made things less responsive. I guess it was this cursor blink + coroutine sync issue or maybe a placebo effect from my side :), testing the patch and so far so good.

OK, I think we're good to go. Should call step only as much as it did before, but will now resume coroutines if in focus at their desired timings.

testing it with lsp now, and for some reason typing is more lagged...

Could be that because it's actually returning when you request it now, you may need to adjust your timeouts to back off a bit, as they'll be getting called how much they should be now, instead of a semi-random amount of times.

Also with this something is causing wheel scrolling major lagging

We may have to adjust some of the timeouts; if they're used to things happening with events, they could be overly aggressive.

Leaving this change here for reference which fixes the lag issue mentioned above which was caused by coroutine.yield() without any timeout parameter as discussed on discord:

diff --git a/data/core/init.lua b/data/core/init.lua
index 00322121..386f0276 100644
--- a/data/core/init.lua
+++ b/data/core/init.lua
@@ -1396,11 +1396,10 @@ local run_threads = coroutine.wrap(function()
           else
             core.threads[k] = nil
           end
-        elseif wait then
+        else
+          wait = wait or 0.0025 -- default to 400 times per second if wait not specified on coroutine.yield()
           thread.wake = system.get_time() + wait
           minimal_time_to_wake = math.min(minimal_time_to_wake, wait)
-        else
-          minimal_time_to_wake = 0
         end
       else
         minimal_time_to_wake =  math.min(minimal_time_to_wake, thread.wake - system.get_time())

So far testing seems to indicate better lsp responsiveness without any further changes.

works fine for me

Weren't there issues with yields of 0 and nil?

It's not an issue, per-se; if you yield 0, you shouldn't wait anything.

The only issue is nil. Which is debatable as to what the appropriate behaviour should be. The current behaviour, even before this PR, technically is a wait of 0.

I'm cool with a default wait (and can implement that, if you like), like jgm said, but that technically would be a change, which I'm not sure if that'd be expected.

OK; so upon review; jgm is correct, there's a slight issue with nil, in that wake times will go negative. This shouldn't actually lead to much difference in behaviour, but I'll make his changes to make it more clear here; though I don't know about the default of blank being non-zero.