[MarcBerliner]

Description

Speeds up model build and IDAKLUSolver set-up for large models (hundreds of thousands to millions of states), where several steps scaled worse than linearly or copied large objects needlessly:

• StateVector stores its slices as index ranges instead of a dense boolean evaluation array over every state. Construction, size, Jacobians and CasADi conversion no longer touch an array the length of the state vector for each node.
• CasADi functions and matrices are shared with pybammsolvers by pointer when it links the same CasADi library as the Python package (idaklu.shares_python_casadi), instead of being serialised and rebuilt. Sparse and dense casadi.DMs are filled from NumPy/SciPy buffers in one copy (to_casadi_dm, CSC and CSR), rather than element by element through SWIG. Without a shared library, the serialised path is used as before.
• IDAKLUSolver no longer builds the rhs-only Jacobian, which only ScipySolver uses (BaseSolver.uses_rhs_jacobian).
• Unstructured meshes compute cell geometry per local vertex and group faces with a lexsort (_group_rows) instead of np.unique(axis=0); hex geometry is about 2.5x faster at 400k cells, with identical volumes.
• stack_inputs concatenates input values with one buffer copy.

Shared functions keep CasADi's reference counting, so the solver owns them for its lifetime and releases them on deletion: 40 repeated set-up and solve cycles of a 38k-state model show no memory growth, in either the shared or the serialised mode.

Benchmarks

Best of 3, IDAKLUSolver() defaults, 1 h discharge; end voltages agree with main to 10 digits.

Model	States	Build (ms) main → PR	Set-up (ms) main → PR
SPM	42	18 → 17	15 → 11
DFN	962	52 → 42	53 → 25
DFN, lumped thermal	963	81 → 64	265 → 116
MPM	1204	45 → 39	72 → 36
DFN, 2D current collectors	48202	458 → 72	1350 → 455

Type of change

• Optimisation, with one breaking change and one deprecation (see CHANGELOG.md): StateVector and StateVectorDot now raise ShapeError for stepped, overlapping or out-of-order slices, and their evaluation_array argument is deprecated and ignored.

Important checks:

Please confirm the following before marking the PR as ready for review:

• No style issues: nox -s pre-commit
• All tests pass: nox -s tests
• The documentation builds: nox -s doctests
• Code is commented for hard-to-understand areas
• Tests added that prove fix is effective or that feature works

[codecov]

Codecov Report

❌ Patch coverage is 96.75325% with 5 lines in your changes missing coverage. Please review.
✅ Project coverage is 98.22%. Comparing base (8c6816d) to head (9f72b8a).

Files with missing lines	Patch %	Lines
.../pybamm/src/pybamm/expression_tree/state_vector.py	92.85%	4 Missing ⚠️
...ackages/pybamm/src/pybamm/solvers/idaklu_solver.py	90.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #5994      +/-   ##
==========================================
- Coverage   98.24%   98.22%   -0.02%     
==========================================
  Files         350      350              
  Lines       36112    36173      +61     
==========================================
+ Hits        35478    35532      +54     
- Misses        634      641       +7     

☔ 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.

[github-actions]

Bencher Report

Project	PyBaMM
Branch	perf/large-model-build-v2
Testbed	bare-metal

⚠️ WARNING: Truncated view!

The full continuous benchmarking report exceeds the maximum length allowed on this platform.

🐰 View full continuous benchmarking report in Bencher

[BradyPlanden]

Thanks!