On Thursday, 11 June 2015 at 07:13:10 UTC, deadalnix wrote:
On Thursday, 11 June 2015 at 07:11:33 UTC, rsw0x wrote:
On Thursday, 11 June 2015 at 07:08:02 UTC, Ola Fosheim Grøstad
wrote:
A ground breaking GC will emerge from the synthesis of the
unsurpassable number of endless GC debates. That is the
sanctimony of meritocracy.
actually making a good GC for D is difficult because the only
type of barrier you can use it hardware protection faults. The
performance dropoff isn't _that_ bad from what I've read in
various papers.
I should have an article up in a few weeks detailing my summer
research project on this.
bye.
It IS that bad, and any paper that tell you otherwise is lying
to you.
after extensive testing I found a hardware protection fault +
mprotect to take 2000nsecs on every hardware I tested, the only
time you need them active is during a concurrent mark and sweep.
I'll find out I guess.
Boehm's GC uses this and regularly kept up(~5-10%) with
essentially all of the top of the line GCs in all the papers I
read.
