Hello everyone,
does someone know whats the state of Diffractor.jl is?
I saw many posts from 2021, saying that it would be soon avaible. The current versioin of Diffractor.jl in the repo has a julia version depencey of 1.10. So I think, this one cant be used atm.
Thanks for your time.
Well, you answered your own question: It depends on Julia 1.10 which is unlikely to be released in the next 9 months, so the current state is “experimental”…
Also the project has seen little to no development during 2022. I of course don’t know the reason for this but to me that looks like a dead repo. Happy to be corrected though. My excitement with respect to autodiff currently lies within Enzyme.
There is some discussion on the current state of Diffractor and Enzyme on the Slack from a few weeks ago Slack
Ultimately, I think it’d be healthier for the community if we just pretended Diffractor didn’t exist and wasn’t coming. People waiting for it to come caused a huge amount of damage and slowdown in the AD space. If it arrives in a usable state sometime in the near future, that’s fantastic, but people shouldn’t be hitching their wagon to that. Even if a timeframe for its arrival was announced, I would really caution people against taking that seriously at this point.
The skills required to move Diffractor forward are in high demand in many other areas of the compiler, and nobody (apparently not even the Diffractor devs) can predict when they’ll be able to be concentrated and focused on Diffractor itself.
This isn’t some failing of the Diffractor devs, it’s just a reality that coupling a next generation AD system so tightly to advances in the compiler is tricky business because there’s many things we want compiler developers to work on.