perf: profile and speed up unit tests

[Jammy2211]

Overview

Some unit tests run slow. This task profiles the PyAutoArray test suite to identify the slowest tests, diagnose bottlenecks, and apply targeted optimizations. The outcome will include a reusable prompt template that can be applied to PyAutoGalaxy and PyAutoLens afterward.

Plan

• Profile the full PyAutoArray test suite to rank all tests by duration
• Identify and categorize the slowest tests by bottleneck type (computation, fixtures, I/O)
• Deep-profile the worst offenders with cProfile to find specific hot functions
• Propose targeted speedups for approval before making changes
• Implement approved optimizations and re-profile to confirm improvement
• Produce a reusable prompt for applying the same process to PyAutoGalaxy and PyAutoLens

Detailed implementation plan

Affected Repositories

• PyAutoArray (primary and only)

Work Classification

Library

Branch Survey

Repository	Current Branch	Dirty?
PyAutoArray	main	untracked files only

Suggested branch: feature/unit-test-profiling

Implementation Steps

1. Run pytest --durations=0 on PyAutoArray to get a ranked list of all test durations
2. Categorize the slowest tests by directory and bottleneck type
3. Profile individual slow tests with cProfile to pinpoint hot functions
4. For each slow test, identify root cause: oversized inputs, redundant setup, expensive computation, unnecessary I/O
5. Present findings with proposed fixes for user approval
6. Implement approved changes, verifying each test still passes
7. Re-run full suite to confirm all tests pass and measure overall speedup
8. Write a reusable prompt template (with lessons learned) for PyAutoGalaxy and PyAutoLens

Key Files

• test_autoarray/conftest.py — 70+ fixtures
• test_autoarray/inversion/ — 36 test files (likely hotspot)
• test_autoarray/structures/ — 26 test files
• test_autoarray/profile_tst.py — existing profiling script

Original Prompt

Click to expand starting prompt

This is a generic source code improvement that applies to all five PyAuto repos.

Basically, some unit test run slow, and this dev work aims to profile the unit tests
and speed them up as much as possible. We dont want to change numerical values on unit
tests where possile, hwoever I am also confident currently that the source code is stable
thus for the slower unit tesrts if there are safe changes we can make to speed them up, then we should do so,
but prompt me on the.

To begin, can we run this on PyAutoArray and I'll see how it goes and we
can then run it on the other repos.

[Jammy2211]

Shipped — 2026-04-06

PR

• perf: speed up unit tests 63% by removing JAX from triangle tests #252

Summary

Profiled the full PyAutoArray test suite and reduced total time from 33.9s to 12.5s (63% faster). The main bottleneck was JAX JIT compilation overhead in triangle unit tests. Switched all triangle tests to numpy equivalents, removed pure-JAX tests covered by integration tests, and reduced oversized test inputs.

Session Notes

• Unit tests should not use JAX — numpy equivalents (ArrayTrianglesNp, CoordinateArrayTrianglesNp) already exist and are much faster.
• Numba JIT compilation is a one-time cost per session (~4.5s) with cache=True — subsequent runs are fast.
• Matplotlib plot tests (~3s) are irreducible — figure construction overhead.
• A reusable prompt for PyAutoGalaxy/PyAutoLens profiling is the next deliverable.