Moritz Schauer

It would help if you explain on a concrete example what cannot be done.

So this is wrong: ``` using Distributions function mh(target, proposal; init=0.0, samples=100_000) x = init xs = [x] for iter in 2:samples xᵒ = rand(proposal(x)) l = logpdf(target, xᵒ) -...

Probably possible to find models, where it works cc @gaurav-arya

I think I would test that `copy` works as intended (same data, different memory)

It appears that this can be done with minimal overhead. Basically, iterators already implement message passing with the message `nothing` for starting and stopping.

The solution can be to define a "PostProcess(state, message)" state which sends `Close()` up-range.

I added an example explaining how this can be done: https://github.com/mschauer/DynamicIterators.jl/blob/master/example/ressourcemanagement.jl

Could this be related to the observation that https://github.com/mschauer/Trajectories.jl/blob/master/src/unroll1.jl will only run if defined in the same module as the caller?

There are two ways to deal with a space homogenous jump distribution when an absorbing boundary comes into play: jumps which formerly would go beyond the boundary do not happen...

Our example had continuous dynamics and only the resampling step was discrete. In your example the resampling step is also discrete. The smoothing backend which will take care of this...