Affected Repositories

• PyAutoFit (primary)

Branch Survey

Suggested branch: feature/fork-context-darwin-test

Implementation Steps

1. In test_autofit/non_linear/test_fork_context.py, add a test that uses monkeypatch.setattr(sys, "platform", "darwin") and asserts fork_context() is multiprocessing and that it has Process, Queue, Pool.
2. Add a test that, on the darwin path, replaces multiprocessing.get_context with a recording spy (monkeypatch) and asserts no call was made with no argument (the pre-fix: do not fix the multiprocessing start method at import time on macOS #1657 import-time call).
3. Keep test_fork_context_is_fork (Linux get_context("fork")) unchanged; monkeypatch restores all state so conftest's fork setting is unaffected.
4. Witness: temporarily revert autofit/non_linear/parallel/context.py final line to return multiprocessing.get_context(); new tests fail; restore; they pass.
5. Run python -m pytest test_autofit/non_linear/test_fork_context.py -q and the full test_autofit suite.

Key Files

• autofit/non_linear/parallel/context.py — fork_context() (code under test, unchanged)
• test_autofit/non_linear/test_fork_context.py — new tests

Links: PyAutoFit#1657, https://github.com/orgs/PyAutoLabs/discussions/27

PyAutoFit: add a regression test for fork_context() not fixing the multiprocessing start…

Type: bug
Target: PyAutoFit
Repos:

• PyAutoFit
  Difficulty: small
  Autonomy: safe
  Priority: high
  Memory: wiki/lensing/sources/dark-matter-substructure.md; wiki/galaxies/sources/massive-ellipticals.md; wiki/galaxies/sources/cosmos-survey.md
  Status: formalised
  Consequence: glance
  Witness: a new test under test_autofit that monkeypatches sys.platform="darwin" and reloads parallel.context asserts multiprocessing.context._default_context._actual_context is None; it fails on the pre-fix: do not fix the multiprocessing start method at import time on macOS #1657 code (revert the context.py hunk locally) and passes on main
  Review-minutes: 3
  Unattended: ready

PyAutoFit: add a regression test for fork_context() not fixing the multiprocessing start method at import on darwin/win32.

Community PR PyAutoFit#1657 (@samlange04, discussion https://github.com/orgs/PyAutoLabs/discussions/27, merged 2026-10-07) changed autofit/non_linear/parallel/context.py so that on non-Linux platforms fork_context() returns the multiprocessing module instead of multiprocessing.get_context(), because the no-argument get_context() call at import time (parallel/process.py:47 builds class Process(fork_context().Process)) locked the interpreter's default start method and made any later set_start_method("spawn") raise "context has already been set". The PR has no test and Ubuntu CI cannot reach the darwin/win32 branch, so the fix is unprotected against regression.

Fix: add a test (next to test_fork_context.py) that monkeypatches sys.platform = "darwin", reloads autofit.non_linear.parallel.context and process, and asserts multiprocessing.context._default_context._actual_context is None afterwards (i.e. importing did not fix the start method), and that the returned object exposes .Process, .Queue and .Pool. Keep the existing Linux assertion (get_context("fork")) intact and restore the module state after the test so conftest's fork setting is unaffected.