[oschulz]

Hi! I maintain ArraysOfArrays.jl, and since v1 is now out I went through the downstream packages that depend on it — this PR makes ForwardBackward.jl work with it.

What changed in ArraysOfArrays v1

The element type of an ArrayOfSimilarArrays is now a view rather than a copy, and as part of that cleanup the nested-array constructors (ArrayOfSimilarArrays(A::AbstractArray{<:AbstractArray})) and nestedview were deprecated in favour of convert/stacked and sliced.

The changes here

Two call sites in src/manifolds.jl use the deprecated APIs:

before	after
ArrayOfSimilarArrays(state)	convert(ArrayOfSimilarArrays, state)
nestedview(nt.data, length(innersize(state.state)))	ArrayOfSimilarArrays{eltype(nt.data), length(innersize(state.state))}(nt.data)

Both replacements are cross-version: convert(::Type{ArrayOfSimilarArrays}, ::AbstractArray{<:AbstractArray}) and the ArrayOfSimilarArrays{T,M}(flat_data) constructor exist, undeprecated, in ArraysOfArrays 0.6.5, 0.6.6 and 1.0.x alike. So the compat entry widens to "0.6.5, 1" rather than dropping 0.6 — nobody currently on 0.6 is forced to move.

Everything else already carried over unchanged: flatview, innersize, and the <:ArrayOfSimilarArrays dispatch in Functors.functor all behave the same in v1.

Nothing else is touched — no version bump, no julia compat change, no test changes.

Verification

Against ArraysOfArrays 1.0.2, 0.6.6 and 0.6.5 in turn, each in a fresh julia --depwarn=yes session:

• Pkg.test() — 93/93 passing on all three.
• The package's own test suite has no Manifolds dependency, so it never touches these two lines. I therefore ran targeted checks with Manifolds loaded (so ManifoldsExt is active): ManifoldState on Sphere(2) from a nested array of points, on ProbabilitySimplex(2) via the integer constructor in the extension, plus Functors.functor/fmap round trips through the reconstruct closure (which is what exercises the second line), tensor, similar, copy, interpolate and endpoint_conditioned_sample (deterministic and stochastic).
• All points stayed on their manifolds, and the recorded numeric outputs are byte-identical across all three ArraysOfArrays versions.
• No deprecation warnings on any version. As a negative control I confirmed the harness does catch both original forms under v1, so the clean run isn't a false negative.

Two small notes

• Your CompatHelper workflow is currently in state disabled_inactivity (GitHub auto-disables scheduled workflows on repos after a period of inactivity), so no automated compat bump would have reached you here — worth re-enabling if you want those PRs back.
• Entirely optional, but a minimal Manifolds-based testset would give CI coverage of this ManifoldState/Functors path, which is the only part of the package that touches these APIs.

A small request

If this looks good, a patch release (0.1.6) would let the registry admit ArraysOfArrays 1 for ForwardBackward — Flowfusion.jl then picks it up automatically through its own bounds. Happy to adjust anything here, and thanks for the package!

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 0% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 49.65%. Comparing base (1fdd4d6) to head (19905f1).

Files with missing lines	Patch %	Lines
src/manifolds.jl	0.00%	2 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main      #23   +/-   ##
=======================================
  Coverage   49.65%   49.65%           
=======================================
  Files           8        8           
  Lines         431      431           
=======================================
  Hits          214      214           
  Misses        217      217           

☔ View full report in Codecov by Harness.
📢 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.