@bpr Given that I spent months on the problem, then gave up and worked on 
`--gc:orc` (with much better progress), it's anything but a "temporary 
implementation detail". But one can write new async implementations that work 
with pure ARC. Whether that's important or not remains unclear, I _might_ have 
discovered an optimization of the ORC algorithm that could turn the overhead of 
the cycle collector to the level of "noise", esp for async where some IO is 
usually involved too. 

Reply via email to