On Thu, May 19, 2022 at 07:07:45AM +1200, rikki cattermole via Digitalmars-d-announce wrote: > > On 19/05/2022 7:03 AM, H. S. Teoh wrote: > > We keep coming back to this impasse: write barriers. It's high time > > somebody implemented this in a dmd fork so that we can actually test > > out more advanced GC designs. > > No. Not dmd. > > LDC or GDC. > > DMD is not suitable for experimentation due to the situation with > atomics. > > Advanced GC's may need concurrent data structures like lock-free, and > you cannot implement them with dmd due to the atomic instructions > being 3 function calls deep. You get segfaults. Been there done that, > what a waste of 7 months.
Sounds good, do it in ldc/gdc, then. Nowadays I only ever use dmd when I need quick turn-around time anyway. In terms of codegen it's pretty lackluster, for production builds my go-to is LDC. T -- Change is inevitable, except from a vending machine.
