On Wednesday, 5 February 2020 at 15:54:23 UTC, Steven
Schveighoffer wrote:
I think we should be more selective about when we make parallel
threads to scan (maybe above a certain size of used memory?).
-Steve
That threshold would have to adapt to each system. How about
delaying thread startup until the first actual mark phase or the
first run after the first 100ms of runtime or so? The main goal
should be to minimize the runtime impact. It doesn't even make
sense to me to run a GC cycle for a program that runs for a
extremely short time only.
