prob fails with multiple chains

[awellis]

I'm not sure whether this is by design, but the prob and logprob macros don't work when sampling with more than one chain:

using Distributions, Turing

@model demo(x) = begin
    m ~ Normal()
    for i ∈ eachindex(x)
      x[i] ~ Normal(m, 1)
    end
end

x = rand(Normal(3, 1), 10)

chn = sample(demo(x), NUTS(), MCMCThreads(), 2000, 2, save_state = true)
logprob"x = 1 | chain = chn"

produces an error

ERROR: DimensionMismatch("tried to assign 2 elements to 1 destinations")

[devmotion]

IMO it's a bug. The underlying problem is the same as in TuringLang/Turing.jl#1301.

[devmotion]

Fixed in DynamicPPL 0.9 and Turing 0.14.