Mastodon Feed
dysfun@treehouse.systems ("gaytabase") wrote:
the only cost i don't like with stackful coroutines is allocating stacks. i have integrated stack watermarking in gnt to help you measure the actual depth your stacks get to in practice, but with compiler support we could do much better than that. just probably not with any of the existing compilers...
otherwise, well spawning an actor is literally under 20 cycles on my machine when branch predicted correctly. this is not your bottleneck and i don't think it takes much polling to make up for it.