RepeatSampler doesn't use MCMCChains.Chains

[penelopeysm]

Minimal working example

julia> using Turing

julia> @model f() = x ~ Normal()
f (generic function with 2 methods)

julia> sample(f(), RepeatSampler(MH(), 3), 1000)
Sampling 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| Time: 0:00:02
1000-element Vector{Turing.Inference.Transition{OrderedDict{AbstractPPL.VarName, Any}, Float64, @NamedTuple{}}}:

Description

blah

Julia version info

Manifest

[penelopeysm]

Maybe this is intended? But it's clearly very difficult to tease out which methods are meant to work on RepeatSampler and which ones aren't.

[penelopeysm]

OK I think it's intended. The root cause is that RepeatSampler doesn't go via DynamicPPL (it doesn't get wrapped in DynamicPPL.Sampler), it just goes straight to AbstractMCMC, and thus DynamicPPL.default_chain_type isn't ever invoked. The solutions would be to either wrap RepeatSampler in Sampler, or to move default_chain_type to AbstractMCMC. But I think it isn't really a problem that needs to be solved.