disable fallback for returned and pointwise_logdensities

[penelopeysm]

It turns out I actually also fixed this in #1130, so this is really just a backport to 0.38.

For an example of the problematic behaviour, run this with current Turing release + DPPL 0.38.9:

using Turing
@model function f()
    return x ~ product_distribution((; a = Normal()))
end
model = f()
chain = sample(model, NUTS(), 10)

chain[Symbol("x.a")][1:5]
#=
 -1.1115679614699299
 -1.1115679614699299
 -1.1115679614699299
 -1.1115679614699299
  0.6485348468368219
=#

rets = returned(model, chain); # will print loads of warnings
rets[1:5]
#=
 (a = 0.43617950918721293,)
 (a = -0.03530447045796848,)
 (a = -0.22527167664558378,)
 (a = -1.9301798082540629,)
 (a = 0.8968461391852851,)
=#

This PR changes it such that the call to returned errors instead of silently giving wrong results.

Note that FlexiChains works correctly (both before and after this PR):

using Turing, FlexiChains
@model function f()
    return x ~ product_distribution((; a = Normal()))
end
model = f()
chain = sample(model, NUTS(), 10; chain_type=VNChain)
chain[@varname(x.a)][1:5]
#=
  0.21193332804635567
 -0.3905216939777102
 -0.1371019215195824
 -0.1371019215195824
 -0.7884639857653091
=#
rets = returned(model, chain);
rets[1:5]
#=
 (a = 0.21193332804635567,)
 (a = -0.3905216939777102,)
 (a = -0.1371019215195824,)
 (a = -0.1371019215195824,)
 (a = -0.7884639857653091,)
=#

[github-actions]

Benchmark Report

• this PR's head: 9df4914b9df7278fc97b106fb74c8c0fa748f26b
• base branch: 052bc1950df3a42e14b56eae51af236881092f90

Computer Information

Julia Version 1.11.7
Commit f2b3dbda30a (2025-09-08 12:10 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 4 × AMD EPYC 7763 64-Core Processor
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, znver3)
Threads: 1 default, 0 interactive, 1 GC (on 4 virtual cores)

Benchmark Results

┌───────────────────────┬───────┬─────────────┬───────────────────┬────────┬─────────────────────────────────┬───────────────────────────┐
│                       │       │             │                   │        │        t(eval) / t(ref)         │     t(grad) / t(eval)     │
│                       │       │             │                   │        │ ──────────┬───────────┬──────── │ ──────┬─────────┬──────── │
│                 Model │   Dim │  AD Backend │           VarInfo │ Linked │      base │   this PR │ speedup │  base │ this PR │ speedup │
├───────────────────────┼───────┼─────────────┼───────────────────┼────────┼───────────┼───────────┼─────────┼───────┼─────────┼─────────┤
│               Dynamic │    10 │    mooncake │             typed │   true │    503.08 │    519.30 │    0.97 │  9.93 │    7.67 │    1.29 │
│                   LDA │    12 │ reversediff │             typed │   true │   3111.80 │   3111.50 │    1.00 │  2.02 │    2.19 │    0.92 │
│   Loop univariate 10k │ 10000 │    mooncake │             typed │   true │ 180300.68 │ 161447.63 │    1.12 │  5.36 │    5.89 │    0.91 │
│    Loop univariate 1k │  1000 │    mooncake │             typed │   true │  14745.62 │  15074.87 │    0.98 │  5.83 │    5.78 │    1.01 │
│      Multivariate 10k │ 10000 │    mooncake │             typed │   true │  34342.66 │  46839.95 │    0.73 │  9.69 │    7.61 │    1.27 │
│       Multivariate 1k │  1000 │    mooncake │             typed │   true │   3666.27 │   5043.36 │    0.73 │  9.21 │    6.97 │    1.32 │
│ Simple assume observe │     1 │ forwarddiff │             typed │  false │     18.63 │     19.05 │    0.98 │  1.88 │    1.88 │    1.00 │
│           Smorgasbord │   201 │      enzyme │             typed │   true │   2814.39 │   3018.01 │    0.93 │  4.75 │    4.88 │    0.97 │
│           Smorgasbord │   201 │ forwarddiff │       simple_dict │   true │  28396.66 │  26062.62 │    1.09 │ 23.21 │   26.69 │    0.87 │
│           Smorgasbord │   201 │ forwarddiff │ simple_namedtuple │   true │   1159.59 │   1314.68 │    0.88 │ 82.63 │   84.88 │    0.97 │
│           Smorgasbord │   201 │ forwarddiff │             typed │  false │   2752.72 │   3040.17 │    0.91 │ 91.77 │   49.27 │    1.86 │
│           Smorgasbord │   201 │ forwarddiff │      typed_vector │   true │   2817.46 │   3083.47 │    0.91 │ 43.35 │   51.23 │    0.85 │
│           Smorgasbord │   201 │ forwarddiff │           untyped │   true │   2516.98 │   2635.01 │    0.96 │ 47.48 │   48.91 │    0.97 │
│           Smorgasbord │   201 │ forwarddiff │    untyped_vector │   true │   2599.08 │   2689.81 │    0.97 │ 44.09 │   49.09 │    0.90 │
│           Smorgasbord │   201 │    mooncake │             typed │   true │   2823.78 │   3031.10 │    0.93 │  5.58 │    4.90 │    1.14 │
│           Smorgasbord │   201 │ reversediff │             typed │   true │   2925.00 │   3203.64 │    0.91 │ 56.86 │   53.55 │    1.06 │
│              Submodel │     1 │    mooncake │             typed │   true │     30.48 │     28.09 │    1.09 │  5.00 │    5.36 │    0.93 │
└───────────────────────┴───────┴─────────────┴───────────────────┴────────┴───────────┴───────────┴─────────┴───────┴─────────┴─────────┘

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 81.58%. Comparing base (052bc19) to head (9df4914).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1159   +/-   ##
=======================================
  Coverage   81.58%   81.58%           
=======================================
  Files          40       40           
  Lines        3845     3845           
=======================================
  Hits         3137     3137           
  Misses        708      708           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[penelopeysm]

This needs tests, but I won't do it today.