Restrict tuple nodeselect methods to integer indices

[ChrisRackauckas]

🚧 UNREVIEWED — awaiting review by @ChrisRackauckas. Filed by an AI agent running as @ChrisRackauckas; Chris has not seen this.
Harness: Claude Code 2.1.300 · Model: claude-opus-5-5
Conversation: https://claude.ai/code/session_01VZjwf7Ux9YtTzwHGHMqnsq

Please ignore this draft until it has been reviewed by @ChrisRackauckas.

Fixes a regression from #172 (merged, unreleased); it should land before the release in #173.

What changed and why

#172 added recursive tuple methods of nodeselect that evaluate i == 1 and i - 1 on any index type. A CartesianIndex first-level index therefore throws MethodError: no method matching -(::CartesianIndex{1}, ::Int64) where it used to return the element. It is reachable through public getindex(m, i, I...) when m.values is empty, e.g. plant[CartesianIndex(2), 1, 1]. Both tuple methods now take i::Integer, so other index types go back to the generic ns[i][I...] method. Both methods need the restriction: restricting only the ::Tuple one makes nodeselect(::Tuple{}, ::Int, ...) ambiguous.

nodeselect(([10, 20], [1.5, 2.5, 3.5]), CartesianIndex(2), 1) returns 1.5 again (as before #172).

Verification

Regression test added to test/tuple_nodes.jl (CartesianIndex first-level index through public getindex on tuple-node structures with empty values, at two depths).

Before (fix stashed, test present, julia +release 1.13.2):

tuple_nodes: Error During Test at test/tuple_nodes.jl:55
  Expression: plant1[CartesianIndex(2), 1, 1] == plant1[2, 1, 1] == 4.1
  MethodError: no method matching -(::CartesianIndex{1}, ::Int64)
tuple_nodes: Error During Test at test/tuple_nodes.jl:56
  Expression: scenario[CartesianIndex(1), 1, 1] === scenario[1, 1, 1]
  MethodError: no method matching -(::CartesianIndex{1}, ::Int64)
Test Summary: | Pass  Error  Total
tuple_nodes   |   21      2     23

After (same file, with the fix), on both 1.13.2 and 1.10.12:

Test Summary: | Pass  Total
tuple_nodes   |   23     23
MultiScaleArrays.nodeselect(([10, 20], [1.5, 2.5, 3.5]), CartesianIndex(2), 1) = 1.5
Test.detect_ambiguities(MultiScaleArrays) = Tuple{Method, Method}[]

The type-stability gain from #172 is kept: the @inferred getindex(community, 4) and @inferred getindex(scenario, 8) lines in test/tuple_nodes.jl pass on 1.13.2 and 1.10.12 (they take an Int index, which still hits the recursive methods).

• GROUP=Core julia +release --project -e 'using Pkg; Pkg.test()' (1.13.2, JULIA_NUM_THREADS=4): Testing MultiScaleArrays tests passed
• GROUP=Core julia +lts --project -e 'using Pkg; Pkg.test()' (1.10.12): Testing MultiScaleArrays tests passed
• GROUP=QA on 1.13.2: passed
• Runic: no changes. typos: clean.

Not verified: Core 32-bit and pre lanes (left to CI). No docs changes (no public API change).

Risk assessment

• Risk: low
• Blast radius: nodeselect (internal) via getindex(m, i, I...) on structures with tuple nodes. Integer indices take the same path as in Make tuple-node getindex type-stable; replace @test_broken @inferred with @inferred #172; non-integer first-level indices take the pre-Make tuple-node getindex type-stable; replace @test_broken @inferred with @inferred #172 generic path. No public API change, no new dependencies.
• Evidence: regression test fails before / passes after (above); GROUP=Core passes on 1.13.2 and 1.10.12; QA passes; no method ambiguities.
• Independent review: codex gpt-6.1-sol rated it low, high confidence: narrow dispatch correction restores generic indexing while preserving tested inference on both supported Julia versions. This rating was selected before reading the author's risk assessment. (review run locally; codex-cli 0.162.1 session 01a142db-f871-7e33-9164-14f6b68fc93b, no published URL)
• Merge: needs human review — fixes an unreleased regression that blocks the release PR; small diff, but it touches indexing.

Independent reviewer's risk assessment (codex-cli 0.162.1 / gpt-6.1-sol, verdict MERGE)

• Risk: low
• Blast radius: Internal tuple-node selection reached through public hierarchical getindex. Integer indices retain recursive selection; other index types use the original generic tuple indexing method. No dependencies or public signatures change.
• Evidence: Reviewed head 4a48db28847e9bac7db6d4bbd2182868fa0c8a89. On Julia 1.13.2 and 1.10.12, the full test/tuple_nodes.jl passed 23/23 assertions, including both deep @inferred checks. Both new CartesianIndex assertions errored with PR 172's methods restored. A 108-case comparison against pre-172 selection produced zero value/exception-type differences on either version. Test.detect_ambiguities(MultiScaleArrays; recursive=false) returned an empty list on both. CI remains pending for several lanes; no failed checks were reported when inspected.
• Independent review: codex gpt-6.1-sol rated it low, high confidence: narrow dispatch correction restores generic indexing while preserving tested inference on both supported Julia versions. This rating was selected before reading the author's risk assessment.
• Merge: auto-merge candidate — small, tested regression repair; wait for required CI and the repository's review requirements.

---

🤖 Posted by an AI agent — harness: Claude Code 2.1.300 · model: claude-opus-5-5
Conversation: https://claude.ai/code/session_01VZjwf7Ux9YtTzwHGHMqnsq

🤖 Generated with Claude Code