On Friday, 24 May 2013 at 10:20:12 UTC, Vladimir Panteleev wrote:
Leandro wasn't using Windows ;)
True but assuming windows isn't completely degenerate the
overhead of a page fault is going to be pretty close to identical
on the same cpu architecture.
On Friday, 24 May 2013 at 11:15:21 UTC, Artur Skawina wrote:
A page fault per every page written to between every GC run + a
user
space callback for every such fault + a page sized copy every
time is
not going to perform very well...
(There are ways to avoid these costs, but no idea if it's
easily doable on
windows...)
artur
It's only between the time a GC run is started and the time it
finishes, exactly the same as in the talk. A generational GC
should be able to do complete mark cycles fairly infrequently.
