I guess technically it's now 2026, but I'm not changing the date on this one https://pvk.ca/Blog/2025/12/30/six-versions-accessing/
| Blog | https://jseigh.wordpress.com/ |
| Code Repository | https://github.com/jseigh |
Joe Seigh
@jwseigh
- 4 Followers
- 10 Following
- 20 Posts
This #github panic where everyone is running around and screaming is almost as entertaining as when Jen broke the internet.
A gold plated prop sledge hammer would have been more apropos but then there would have been the inevitable tweet "WORST PUTTER EVER!!!" so there's that.
I can do a more detailed write up of this and the other wait-free hazard pointer techniques but it definitely won't fit into 500 chars.
There are about 3 or 4 ways to make #hazardpointers #waitfree instead of merely #lockfree. I had though Linux restartable sequences (RSEQ) couldn't be inlines but apparently they can. I did a quick and dirty perf rest and it looks like protect() runs 3x faster w/o that conditional branch. But this only works on Linux and you need asynchronous memory barriers.
Found my old Usenet post on using a queue to bypass syscalls, kind of like #io_uring
https://groups.google.com/g/comp.arch/c/32oorU6fTts/m/d-dYChmSosEJ
Test post to see if this postage stamp sized panel is usable. I might need assistance from Alfur.