Will Tebbutt

Lovely, thanks for this. Is there a recommended way to test that my overloads are implemented correctly?

Haha excellent -- thanks for the info!

Thanks for having a look at this! > 1. Dealing with the RNG will be the user's responsability. Before Does this have any implications for integration with Turing.jl? i.e. does...

Thanks for the data point. Essentially the final item on my todo list is sorting out various type inference issues in the current implementation. Once they're done, we should see...

> @willtebbutt running models against this PR I see a large performance drop: @FredericWantiez I'm finally looking at sorting out the performance of the Libtask updates. I'm struggling to replicate...

But, additionally, the latest version of the PR should address the various performance issues we previously had. There is one important change though: you need to pass a type to...

Very happy to give this a review -- when you're ready for one, please do tag me.

I'm not sure that @mhauru is suggesting we could make the entire thing allocation-free, just specific pieces that we're confident ought not to allocate. I agree that making the whole...

I fully agree with your thoughts on this. The performance gains could be very substantial.

Assuming that we're not considering the `\ell` field part of the public interface, this looks like an issue in Turing.jl -- it assumes [here](https://github.com/TuringLang/Turing.jl/blob/29a134245b2499d59fa992420eba37ab2b9f5945/src/mcmc/abstractmcmc.jl#L84) that the `\ell` field of the...