Nenad Rakocevic
Nenad Rakocevic
Can't reproduce the issue with latest Red: ``` generate array NO-COPY: cycle # 1 Time: 0:00:06.43293 generate array NO-COPY: cycle # 2 Time: 0:00:21.193 generate array NO-COPY: cycle # 3...
@hiiamboris I suspect something is wrong with the implementation, it should only take a fraction of that time to free the memory. After doing some tests in debug mode (to...
> BTW, I ran this test in Rebol2, seems the recycle does nothing, no memory usage reduced. Me too, same result, Rebol2 does not release the memory. R3 does a...
> Let me broaden the issue a bit. Not about fixing this GC, but about what we'll have in 1.0. We'll have the current GC, with some extra fixes and...
> We'll still have to scan the whole "big series" to mark series that appear in it, no? It has to depend on the size of the series too, not...
> E.g. if I allocate a 1TB series, GC won't ever need to scan it. Maybe it's just integers anyway? Maybe I've missed something in your proposals, but how do...
Seems too expensive and not scalable. A series can have an unlimited number of references. Having one hashtable for each node and tracking all the changes would have a significative...
Seems over-complicated... "O(number of series) vs O(amount of live RAM)." I don't understand how you differentiate those two?
> With scanning we have N nodes (series) by average length M, resulting in O(MN=amount of live RAM) scanning coverage. We don't scan every byte in a series, so it's...
@qtxie I've taken care of the Windows part, I let you process the GTK part.