David Widmann

I guess the name is motivated by the fact that it contains much more than just the samples. I think `Sample(model)` or `Draw(model)` raises the question which sampler is used....

> ATM, SimpleVarInfo contains only the samples and the log-probability (which I think of as part of the sample). You mentioned `VarInfo` and hence I referred to it. Both the...

> That would also be good; no reason not to do both! The main reason to not do both is that it creates confusion if there are multiple official ways...

> maybe just change SimpleVarInfo then? I'm not sure if we want users to work with `SimpleVarInfo` explicitly. Most functionality (apart from the sampling?) works with just providing named tuples...

> but people interested in adding their own samplers to Turing, for instance. I meant them as well. As the MH example shows, they don't have to deal with `SimpleVarInfo`....

xref: https://github.com/TuringLang/DynamicPPL.jl/pull/360#issuecomment-1038283245

Speecifying both a sampling context and an RNG is redundant, hence IMO this should not be supported. But as mentioned in the comment linked above, we should clean the dispatches...

> Ahh, ok, looks like it requires either all the arguments to be passed or none of them. No, there are a bunch of different constructors: https://github.com/TuringLang/DynamicPPL.jl/blob/aeb5e03fdfe6958e4f79d9ccc5b01b84b93371ff/src/contexts.jl#L135-L138 You can pass...

> Support for Gibbs sampling: e.g. GID stuff in VarInfo IIRC we don't use GID etc in the Gibbs sampler anymore.

I think that's done solely by the sub-samplers and their parameter space (eg https://github.com/TuringLang/Turing.jl/blob/9f8a9c4c476095d45246b4924d5cd542f3f8d506/src/inference/gibbs.jl#L243)? The logic was changed quite a bit in https://github.com/TuringLang/Turing.jl/pull/1500.