Path to Julia 1.11

[penelopeysm]

• Update Manifest, make sure to run with Julia 1.11
• Dependency resolution failed offline because Pathfinder.jl doesn't yet have a version compatible with Turing 0.35. I bumped the compat entry on a local clone of Pathfinder, which allowed dependencies to be successfully resolved (which implies that there aren't any other packages that also have conflicting dependency resolution). I opened Bump Turing/DynamicPPL compat mlcolab/Pathfinder.jl#214; when this is merged and released we can update the deps.
• Precompilation failed. Remove BijectorsEnzymeExt on 1.11.1 Bijectors.jl#337 should fix it. Get this merged and released.
  • Bump Bijectors compat on Turing Bump Bijectors to 0.14.0 Turing.jl#2384
    • Bump Bijectors compat on AdvancedVI Add compatibility for Bijectors 0.14 AdvancedVI.jl#136
    • Bump Bijectors compat on DynamicPPL Bump Bijectors version to 0.14 DynamicPPL.jl#701
      • Get DynamicPPL CI passing with Turing
  • Make Bijectors compatible with Enzyme 0.13 Explicit Enzyme rules on Enzyme 0.13 Bijectors.jl#350 This is necessary because AdvancedVI requires Enzyme 0.13.

[yebai]

This is necessary because AdvancedVI requires Enzyme 0.13 and Bijectors isn't compatible with it

Isn't Enzyme a weakdeps for both AdvancedVI and Bijectors?

[penelopeysm]

SciMLSensitivity has Enzyme as a strong dep (so this is the downside of having a single environment for docs, #497).

[yebai]

I added a comment here. Let's hope that SciMLSensitivity can move towards package extension to support autograd libraries.

[ChrisRackauckas]

SciMLSensitivity has Enzyme as a strong dep (so this is the downside of having a single environment for docs, #497).

Let's hope that SciMLSensitivity can move towards package extension to support autograd libraries.

Why do you think that would solve the problem? Even if we did have extensions, Enzyme would still be a strong dep, right? I don't understand the logic here. There's no way to call into Enzyme automatically if it's not made a dep.