On Friday, 11 September 2015 at 01:04:56 UTC, Brandon Ragland
wrote:
There was a lovely article by a fellow for his PhD on how D
garbage collector was literally killing his JavaScript engine,
It was a she :-).
Some people have focused on the GC, but you actually need to
change the strategy so you look at fewer cachelines, prefetch
perfectly and do less work at runtime. I would estimate that you
could improve global mark-sweep by 2-4x, but it takes a lot of
work, not only on the runtime. And maybe 2-4x improvement is too
little anyway.
And @nogc is just a band-aid fix. Might as well go back to C or
C++ and leave the silly @nogc behind with all it's weird
integration rules when working around managed memory.
I don't agree. I think D could find a sweet-spot as a cleaned up
C/C++ solution, but then you need to change the semantics
somewhat.
I'm very much in favour of a reduced version of D that keeps, and
improves, on the compile time advantage D has. Essentially is a
zero-runtime language that is better than C.
