InitContext, part 5 - Remove SamplingContext, SampleFrom{Prior,Uniform}, {tilde_,}assume

[penelopeysm]

Part 1: Adding hasvalue and getvalue to AbstractPPL
Part 2: Removing hasvalue and getvalue from DynamicPPL
Part 3: Introducing InitContext and init!!
Part 4: Using InitFromParams to implement predict, returned, and initialize_values

This is part 5/N of #967.

This PR removes everything that is no longer needed. SamplingContext, SampleFromPrior, SampleFromUniform, now have direct one-to-one replacements (albeit with slightly different behaviour since they now always overwrite variables in the varinfo).

It also removes assume and tilde_assume.

Prior to this PR we had two different kinds of assume, one with a sampler and one without. Now we only have the one without, so we can just move that definition into tilde_assume!!(::DefaultContext, ...).

Finally, tilde_assume has been subsumed into tilde_assume!! as we can just dispatch on the type of right. (Previously this wasn't possible because there was a lot of stuff about is_rhs_model, etc. etc. which was removed in #960.)

Closes #859
Closes #955

[github-actions]

Benchmark Report for Commit f4e5f4b

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

|                 Model | Dimension |  AD Backend |      VarInfo Type | Linked | Eval Time / Ref Time | AD Time / Eval Time |
|-----------------------|-----------|-------------|-------------------|--------|----------------------|---------------------|
| Simple assume observe |         1 | forwarddiff |             typed |  false |                  9.0 |                 1.6 |
|           Smorgasbord |       201 | forwarddiff |             typed |  false |                679.8 |                46.1 |
|           Smorgasbord |       201 | forwarddiff | simple_namedtuple |   true |                323.9 |                74.8 |
|           Smorgasbord |       201 | forwarddiff |           untyped |   true |               1227.4 |                30.1 |
|           Smorgasbord |       201 | forwarddiff |       simple_dict |   true |               6910.3 |                29.8 |
|           Smorgasbord |       201 | reversediff |             typed |   true |               1535.0 |                29.0 |
|           Smorgasbord |       201 |    mooncake |             typed |   true |               1075.6 |                 4.6 |
|    Loop univariate 1k |      1000 |    mooncake |             typed |   true |               5944.2 |                 4.3 |
|       Multivariate 1k |      1000 |    mooncake |             typed |   true |               1037.0 |                 9.0 |
|   Loop univariate 10k |     10000 |    mooncake |             typed |   true |              66370.1 |                 4.0 |
|      Multivariate 10k |     10000 |    mooncake |             typed |   true |               9192.3 |                 9.7 |
|               Dynamic |        10 |    mooncake |             typed |   true |                136.1 |                11.8 |
|              Submodel |         1 |    mooncake |             typed |   true |                 12.9 |                 5.1 |
|                   LDA |        12 | reversediff |             typed |   true |               1205.6 |                 3.4 |

[codecov]

Codecov Report

❌ Patch coverage is 90.90909% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 82.13%. Comparing base (6d43231) to head (f4e5f4b).
⚠️ Report is 3 commits behind head on breaking.

Files with missing lines	Patch %	Lines
src/context_implementations.jl	88.23%	2 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff              @@
##           breaking     #985      +/-   ##
============================================
+ Coverage     81.20%   82.13%   +0.92%     
============================================
  Files            39       39              
  Lines          3910     3811      -99     
============================================
- Hits           3175     3130      -45     
+ Misses          735      681      -54     

☔ 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 #985 is available at:
https://TuringLang.github.io/DynamicPPL.jl/previews/PR985/

[penelopeysm]

I couldn't find a way to reproduce this test without SamplingContext and the whole tilde-pipeline machinery that this PR removes, so I think that this is no longer relevant.

[penelopeysm]

Actually just for posterity I did find a way to reproduce it when I tried to remove eltype etc. in #1015. But without the changes from that PR I couldn't repro it

[penelopeysm]

sorry this file's diff is a bit of a mess, there are no real code changes, it's just:

• merged tilde_assume, tilde_assume!!, and assume into a single function (they all just called each other)
• added types to the arguments
• added proper docstrings

[mhauru]

Enjoyed this one a lot.

[mhauru]

Is it time to start a HISTORY.md entry? Might be easier to do it here were you can cross-check against what's being removed, rather than once everything is in breaking in a huge diff.

[penelopeysm]

Yeah, good idea to do it in this PR. I'll write one up later and ping you again

[mhauru]

To confirm that I understand how this will play out in Turing.jl: The idea is that samplers don't need to modify the behaviour of the tilde pipeline any more, and thus SamplingContext can go in its entirety, and we don't need things like tilde_assume without !! or assume. And the few that do still need to do that (Gibbs, maybe particles, hopefully nothing else) need to define their own context. Is that right?

[penelopeysm]

I think so, although I'm not 100% sure. We'll probably have to do a similar process to what we did last time for accumulators: make a Turing PR that builds against the breaking branch of DPPL. Things might break. But I'm hoping it won't be too bad haha.

[penelopeysm]

Mildly tangential: I kind of find it weird that we do **Breaking changes** and then ### Subheading because the latter is bigger than the former. I feel like this was originally done to make JuliaRegistrator happy, but I checked again and it claims (e.g. JuliaRegistries/General#139062 (comment)) that it doesn't need to be **Breaking changes**, it can be ## Breaking changes. I'm a bit kind of tempted to reformat the changelog to make it look nicer (not in this PR).

[mhauru]

I, too, have disliked the bolding taking precedence over headings. Happy for that to be changed.

[mhauru]

enables the case where different variables are to be initialised from different sources.

Does this allow different sources in some broader sense than some having fixed values and others using fallback? Like could I do InitFromPrior for some and InitFromUniform for others?

[penelopeysm]

That thought did pop into my mind when I was writing, which led me to mention the custom initialisation strategy just below this. You'd have to write your custom InitFromFoo, but it would be possible in that function to check e.g. the varname and dispatch to prior / uniform accordingly.

[penelopeysm]

Day N of coveralls failing randomly...