Drop the redundant axes(::SparsityPatternCSC, ::Integer) method (invalidates 2k+ ModelingToolkit instances on load)

[ChrisRackauckas-Claude]

What

Removes Base.axes(S::SparsityPatternCSC, d::Integer) = Base.OneTo(size(S, d)) from src/graph.jl. SparsityPatternCSC <: AbstractMatrix{Bool} and defines size, so Base's fallback axes(A::AbstractArray, d) already returns exactly OneTo(size(S, d)) (and OneTo(1) for d > 2, same as before). Adds axes(A) == axes(S) and the d = 3 case to the existing size testset so the behaviour stays pinned.

Why

Adding that method at load time invalidates every axes(A, d) call site that was compiled earlier with an abstract A (axes(::AbstractArray, ::Int64), axes(::AbstractMatrix, ::Int64), axes(::AbstractArray, ::Any) all take a backedge to the method table). SparseMatrixColorings is loaded by OrdinaryDiffEqDifferentiation, i.e. after ModelingToolkit, and on that stack the insertion kills 3,055 cached instances, 2,049 of them in ModelingToolkit / ModelingToolkitBase / ModelingToolkitTearing / Symbolics: __mtkcompile, _mtkcompile, maybe_build_initialization_problem, process_SciMLProblem, alias_elimination!, find_eq_solvables!, ODEProblem, NonlinearProblem. In practice using ModelingToolkit, OrdinaryDiffEq discards the model-building code that ModelingToolkit's precompile workload just cached.

Measured with SnoopCompileCore.@snoop_invalidations using OrdinaryDiffEq after using ModelingToolkit (Julia 1.12.7, ModelingToolkit 11.40, OrdinaryDiffEq 7.8.1, SparseMatrixColorings 0.4.27 vs this branch Pkg.developed):

	instances invalidated by the axes insertion	of which ModelingToolkit-family	build the workload models after using OrdinaryDiffEq
0.4.27	3,055	2,049	12.3 s / 3.4 s
this branch	– (method gone)	0	0.37 s / 0.10 s

(The last column is mtkcompile + ODEProblem for a small ODE and a small DAE that ModelingToolkit's own workload compiles; with the invalidation those take 12.3 s and 3.4 s on first use even though they are in the pkgimage.)

Tests

Pkg.test() on Julia 1.12.7: SparseMatrixColorings | 50464 pass, 31 broken (pre-existing) | 11m44s, tests passed. JuliaFormatter (BlueStyle) reports both touched files formatted.

Context (analysis and scripts): https://github.com/ChrisRackauckas/InternalJunk/issues/88#issuecomment-5537967759. SciML also carries a defensive canary for this pattern in SciML/CommonWorldInvalidations.jl#45; this PR is the fix at the source.

🤖 Generated with Claude Code (model: claude-fable-5-1) on behalf of Chris Rackauckas
https://claude.ai/code/session_01FeJYXni9MkJhPFA9yxYvfn

[ChrisRackauckas]

Seems like an easy win.

[gdalle]

Agreed, docs failure is unrelated

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.20%. Comparing base (e8d80d6) to head (37cb48d).
⚠️ Report is 3 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #338      +/-   ##
==========================================
- Coverage   99.20%   99.20%   -0.01%     
==========================================
  Files          21       21              
  Lines        2151     2150       -1     
==========================================
- Hits         2134     2133       -1     
  Misses         17       17              

☔ 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.