make MCMCThreads etc. respect check_model=false (#2721)

Prior to this PR, if you called

```julia
sample(model, spl, MCMCThreads(), N, n; check_model=false)
```

the model checking here


https://github.com/TuringLang/Turing.jl/blob/ed7f76c7221a756a390d3168ed6e1fcb6f95d263/src/mcmc/abstractmcmc.jl#L121

would be correctly skipped. The problem is that this calls
`AbstractMCMC.mcmcsample`, which then calls the single-threaded
`sample`, but the `check_model` argument is not passed down, so it
defaults to `true` which then checks the model(!!)

This PR fixes it by simply passing `check_model=false` through. (The
point of explicitly setting it to false is that if the user wanted the
check, it would be done on the line above already, there's no need to
check the model `N+1` times.)

The added test fails on main and passes on this PR.

Author: penelopeysm

HISTORY.md

@@ -1,3 +1,7 @@
+# 0.41.4
+
+Fixed a bug where the `check_model=false` keyword argument would not be respected when sampling with multiple threads or cores.
+
 # 0.41.3
 
 Fixed NUTS not correctly specifying the number of adaptation steps when calling `AdvancedHMC.initialize!` (this bug led to mass matrix adaptation not actually happening).

Project.toml

@@ -1,6 +1,6 @@
 name = "Turing"
 uuid = "fce5fe82-541a-59a6-adf8-730c64b5f9a0"
-version = "0.41.3"
+version = "0.41.4"
 
 [deps]
 ADTypes = "47edcb42-4c32-4615-8424-f2b9edc5f35b"

src/mcmc/abstractmcmc.jl

@@ -131,6 +131,7 @@ function AbstractMCMC.sample(
         N,
         n_chains;
         chain_type,
+        check_model=false, # no need to check again
         initial_params=map(_convert_initial_params, initial_params),
         kwargs...,
     )

test/mcmc/abstractmcmc.jl

@@ -6,6 +6,23 @@ using Random: AbstractRNG
 using Test: @test, @testset, @test_throws
 using Turing
 
+@testset "Disabling check_model" begin
+    # Set up a model for which check_model errors.
+    @model f() = x ~ Normal()
+    model = f()
+    Turing.Inference._check_model(::typeof(model)) = error("nope")
+    # Make sure that default sampling does throw the error.
+    @test_throws "nope" sample(model, NUTS(), 100)
+    @test_throws "nope" sample(model, NUTS(), MCMCThreads(), 100, 2)
+    @test_throws "nope" sample(model, NUTS(), MCMCSerial(), 100, 2)
+    @test_throws "nope" sample(model, NUTS(), MCMCDistributed(), 100, 2)
+    # Now disable the check and make sure sampling works.
+    @test sample(model, NUTS(), 100; check_model=false) isa Any
+    @test sample(model, NUTS(), MCMCThreads(), 100, 2; check_model=false) isa Any
+    @test sample(model, NUTS(), MCMCSerial(), 100, 2; check_model=false) isa Any
+    @test sample(model, NUTS(), MCMCDistributed(), 100, 2; check_model=false) isa Any
+end
+
 @testset "Initial parameters" begin
     # Dummy algorithm that just returns initial value and does not perform any sampling
     abstract type OnlyInit <: AbstractMCMC.AbstractSampler end