I have a couple of token count reducer functions:

function poke_map(loopx,loopy,offsetx,offsety)
	poke(0x5f38,loopx)
	poke(0x5f39,loopy)
	poke(0x5f3a,offsetx)
	poke(0x5f3b,offsety)
end

function unpack_split(s)
	return unpack(split(s))
end

As of 0.2.3, the latter one of these, especially when combined with the former, as in

poke_map(unpack_split"4,4,64,32")

has started to cause some weird behaviour in cases where cpu usage is high (around 80% or 90%).

The most obvious symptom is that actions triggered by a simple global timer (i.e. t += 1 in _update60()) start to occur prematurely when the unpack_split() function is called repeatedly in successive _draw() calls.

This sounds complicated enough that it would probably be helpful if you uploaded a full cart.

Also, just to confirm, does the bug go away if you change

poke_map(unpack_split"4,4,64,32")

back to

poke_map(4,4,64,32)

?

Yes, the bug indeed goes away when poke_map() is used with "normal" parameters, as in your example.

In addition, it seems at the moment that using unpack_split() with built-in Pico-8 functions like circ() or rectfill() does not cause this behaviour, but this is not yet fully confirmed.

I'll see to uploading a snippet that reproduces the bug accurately, thanks.