Add more LDF accessor functions that Turing requires

[penelopeysm]

Working on TuringLang/Turing.jl#2790 made me realise that LDF simply doesn't make enough of its internals explicit. Turing thus faces a choice of either depending on DPPL internals or using hacky workarounds. This PR therefore exposes a lower-level constructor for LogDensityFunction that allows Turing to essentially update the model in an LDF without having to rely on private fields (bad) or recreate any of the other arguments (hacky). See TuringLang/Turing.jl#2790 (comment).

[github-actions]

Benchmark Report

• this PR's head: 20c394c1cc9e023e7577ea7c6d4089b35f10c0e3
• base branch: 10380bf7da86efc23768c977988306d4b6f1be41

Computer Information

Julia Version 1.11.9
Commit 53a02c0720c (2026-02-06 00:27 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)      │        t(grad) / t(ref)         │
│                       │       │             │        │ ─────────┬──────────┬──────── │ ───────┬─────────┬──────── │ ──────────┬───────────┬──────── │
│                 Model │   Dim │  AD Backend │ Linked │     base │  this PR │ speedup │   base │ this PR │ speedup │      base │   this PR │ speedup │
├───────────────────────┼───────┼─────────────┼────────┼──────────┼──────────┼─────────┼────────┼─────────┼─────────┼───────────┼───────────┼─────────┤
│               Dynamic │    10 │    mooncake │   true │   274.56 │   295.58 │    0.93 │   7.73 │    7.45 │    1.04 │   2121.23 │   2201.34 │    0.96 │
│                   LDA │    12 │ reversediff │   true │  2578.63 │  2700.81 │    0.95 │   2.00 │    2.10 │    0.95 │   5151.29 │   5673.16 │    0.91 │
│   Loop univariate 10k │ 10000 │    mooncake │   true │ 30728.69 │ 31209.28 │    0.98 │   6.52 │    7.25 │    0.90 │ 200244.12 │ 226160.46 │    0.89 │
├───────────────────────┼───────┼─────────────┼────────┼──────────┼──────────┼─────────┼────────┼─────────┼─────────┼───────────┼───────────┼─────────┤
│    Loop univariate 1k │  1000 │    mooncake │   true │  3128.91 │  4107.06 │    0.76 │   6.38 │    5.55 │    1.15 │  19949.09 │  22805.87 │    0.87 │
│      Multivariate 10k │ 10000 │    mooncake │   true │ 31811.95 │ 40567.63 │    0.78 │   9.82 │    7.99 │    1.23 │ 312447.48 │ 324142.18 │    0.96 │
│       Multivariate 1k │  1000 │    mooncake │   true │  3360.48 │  4351.41 │    0.77 │   9.34 │    7.68 │    1.22 │  31379.69 │  33399.08 │    0.94 │
├───────────────────────┼───────┼─────────────┼────────┼──────────┼──────────┼─────────┼────────┼─────────┼─────────┼───────────┼───────────┼─────────┤
│ Simple assume observe │     1 │ forwarddiff │  false │     0.88 │     0.92 │    0.95 │  10.49 │   11.16 │    0.94 │      9.18 │     10.31 │    0.89 │
│           Smorgasbord │   201 │ forwarddiff │  false │   955.94 │   972.69 │    0.98 │ 143.73 │   77.04 │    1.87 │ 137398.79 │  74939.16 │    1.83 │
│           Smorgasbord │   201 │      enzyme │   true │  1273.65 │  1322.21 │    0.96 │   4.95 │    4.68 │    1.06 │   6300.95 │   6194.39 │    1.02 │
├───────────────────────┼───────┼─────────────┼────────┼──────────┼──────────┼─────────┼────────┼─────────┼─────────┼───────────┼───────────┼─────────┤
│           Smorgasbord │   201 │ forwarddiff │   true │  1342.43 │  1353.30 │    0.99 │  65.52 │   70.19 │    0.93 │  87957.05 │  94990.78 │    0.93 │
│           Smorgasbord │   201 │    mooncake │   true │  1275.64 │  1369.30 │    0.93 │   4.65 │    4.50 │    1.03 │   5934.64 │   6166.51 │    0.96 │
│           Smorgasbord │   201 │ reversediff │   true │  1273.15 │  1327.38 │    0.96 │ 126.94 │  126.76 │    1.00 │ 161611.38 │ 168256.19 │    0.96 │
├───────────────────────┼───────┼─────────────┼────────┼──────────┼──────────┼─────────┼────────┼─────────┼─────────┼───────────┼───────────┼─────────┤
│              Submodel │     1 │    mooncake │   true │     0.88 │     0.92 │    0.95 │  26.41 │   28.90 │    0.91 │     23.12 │     26.70 │    0.87 │
└───────────────────────┴───────┴─────────────┴────────┴──────────┴──────────┴─────────┴────────┴─────────┴─────────┴───────────┴───────────┴─────────┘

[codecov]

Codecov Report

❌ Patch coverage is 96.15385% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 82.35%. Comparing base (10380bf) to head (20c394c).
⚠️ Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
src/logdensityfunction.jl	96.15%	1 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1366   +/-   ##
=======================================
  Coverage   82.35%   82.35%           
=======================================
  Files          49       49           
  Lines        3502     3502           
=======================================
  Hits         2884     2884           
  Misses        618      618           

☔ 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.

[github-actions]

DynamicPPL.jl documentation for PR #1366 is available at:
https://TuringLang.github.io/DynamicPPL.jl/previews/PR1366/