When running a model with programmatically error, samplers like SMC suspends silently without throwing the error.
Simplest example:
julia> using Turing
julia> @model test(xs) = begin
print(xs[2])
x ~ Normal(0, 1)
end
test (generic function with 2 methods)
julia> mf = test([1])
[ Info: Assume - `x` is a parameter
test_model (generic function with 4 methods)
julia> sample(mf, SMC(10)) #=> suspend
TuringLang/Libtask#7