DynamicPPL -> 0.29; Julia -> 1.10; Tapir -> Mooncake

[penelopeysm]

This PR bumps DynamicPPL to 0.29.

Julia

This necessitates dropping support for Julia < 1.10

• the last version of SciMLBase to support Julia 1.9 is 2.24.0
  • which in turn requires ADTypes 0.2
• but DynamicPPL 0.28.3 (and by extension 0.29) requires ADTypes 1.0 (https://github.com/TuringLang/DynamicPPL.jl/pull/638)

Mooncake

Likewise, I've replaced Tapir -> Mooncake as only the latter has true support for DynamicPPL 0.29.

[codecov[bot]]

Codecov Report

Attention: Patch coverage is 72.72727% with 6 lines in your changes missing coverage. Please review.

Project coverage is 86.39%. Comparing base (24e6870) to head (9fb9e95). Report is 1 commits behind head on master.

Files with missing lines	Patch %	Lines
src/mcmc/mh.jl	70.00%	6 Missing :warning:

Additional details and impacted files

@@             Coverage Diff             @@
##           master    #2341       +/-   ##
===========================================
+ Coverage    0.00%   86.39%   +86.39%     
===========================================
  Files          22       22               
  Lines        1533     1573       +40     
===========================================
+ Hits            0     1359     +1359     
+ Misses       1533      214     -1319     

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

[coveralls]

Pull Request Test Coverage Report for Build 11466651176

Details

• 16 of 22 (72.73%) changed or added relevant lines in 3 files are covered.
• 1 unchanged line in 1 file lost coverage.
• Overall coverage increased (+86.4%) to 86.395%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
src/mcmc/mh.jl	14	20	70.0%
<!--	Total:	16	22

Files with Coverage Reduction	New Missed Lines	%
src/stdlib/RandomMeasures.jl	1	33.33%
<!--	Total:	1

Totals
Change from base Build 11314008798:	86.4%
Covered Lines:	1359
Relevant Lines:	1573

---

💛 - Coveralls

[penelopeysm]

Tests still aren't passing, there are still a few changes that need to be made to bring it up to date with https://github.com/TuringLang/DynamicPPL.jl/pull/575.

I don't mind handling this, but it will take me a long time, so if anybody else feels like expediting it please feel free to jump in:)

[mhauru]

I'll ask around if anyone objects to dropping <1.9.

[penelopeysm]

In the meantime I can work to get the PR ready to merge, we will need it at some point anyway whether it's sooner or later:)

[penelopeysm]

As it turns out, Julia 1.9 won't work either, because SciMLBase.

(see e.g. https://github.com/TuringLang/Turing.jl/actions/runs/11153561181/job/31001300867)

[penelopeysm]

I think everything almost works, except for a Mooncake-or-further-upstream bug which I'll report tomorrow.

Edit: https://github.com/compintell/Mooncake.jl/issues/283

[penelopeysm]

@yebai / @mhauru Since we're making fairly sweeping changes here, is it an opportune time to remove Tracker support as well? https://github.com/TuringLang/Turing.jl/issues/2356

[penelopeysm]

@torfjelde Could I get you to take a look at my changes to src/mcmc please? The tests all pass, but I'm a little anxious 😄

Also, this PR fixes the following bug, but not entirely sure what's the best way to write a test for it. (Other samplers already work correctly with this, it's just MH that fails.) Would it make sense to extend test/test_utils/models.jl and test/test_utils/numerical_tests.jl?

julia> using Turing

julia> @model f() = x ~ LKJCholesky(2, 1)
f (generic function with 2 methods)

julia> sample(f(), MH(), 100)
ERROR: MethodError: no method matching length(::LinearAlgebra.Cholesky{Float64, Matrix{Float64}})
[...]

[torfjelde]

(Other samplers already work correctly with this, it's just MH that fails.) Would it make sense to extend test/test_utils/models.jl and test/test_utils/numerical_tests.jl?

I'd just add it to the MH specific tests for now :+1:

[yebai]

@yebai / @mhauru Since we're making fairly sweeping changes here, is it an opportune time to remove Tracker support as well? https://github.com/TuringLang/Turing.jl/issues/2356

It sounds good to me.

[penelopeysm]

This upgrade is also blocked by https://github.com/TuringLang/docs/issues/521 – ideally we want to make sure any relevant docs build with Mooncake. Because we don't want to upgrade to 0.35.0, patch Mooncake to work for the docs, and then release 0.35.1 with a new compat entry for Mooncake just so that the docs work.

However, the Bayesian NN example errors with Mooncake. Will create specific upstrea issues etc when I get the time

[penelopeysm]

Once tests pass (hopefully soon 🙏), still TODO:

• [x] Update API docs
• [x] Changelog note

[penelopeysm]

CI is finally passing, so it's time to ask for you all to take a final look!

To be honest, there are no major code changes since the last round of reviews. Most of the CI errors were related to upstream issues which have since been fixed (mostly due to @willtebbutt's heroic efforts 🙌 ). Julia 1.11 coming out while this PR was open definitely also made it more fun 😄

There are a few outstanding TODOs which I've opened issues for:

• #2369
• #2371
• #2372

However, this PR makes some sweeping changes to the library. To protect this PR from scope creep, I think we should address these in separate PRs. Notably, all of these issues are about tests, not the library code itself.

[yebai]

Great work. Many thanks, @penelopeysm and all!

[sunxd3]

great work! thanks guys