Give each TestScheduler test its own scheduler and project directory

[calvinp0]

TestScheduler does not pass reliably on main. Measured over 12 runs of pytest arc/scheduler_test.py::TestScheduler -n 4 --dist worksteal: 0 clean, 12 failing. The whole file failed 5 of 5.

The failures rotate, which is why this reads as noise rather than a bug: test_check_rxn_e0_by_spc (7 of 12), test_initialize_output_dict (2), test_conformers (1), plus ERROR cascades of up to 11 tests at once.

The cleanest demonstration

$ pytest arc/scheduler_test.py::TestScheduler::test_initialize_output_dict     # on main
1 failed        AssertionError: False is not true

That test only passes when something else runs first. Its opening line asserted that the shared output dict already contained information — true only because an earlier test had dirtied it, and testing nothing about the method under test. It now writes that state itself.

Two causes, not one

1. A shared project directory deleted by tearDownClass. Every fixture used ARC_PATH/Projects/arc_project_for_testing_delete_after_usage3, and tearDownClass rmtree'd it. --dist worksteal can split a single TestCase class across workers, so each worker independently runs setUpClass and tearDownClass — worker A's teardown deletes the tree worker B is still writing into. This is the whole ERROR cascade, and it explains test_check_rxn_e0_by_spc: PIL writing ts_guesses.png into a directory that had just vanished.

2. Tests mutating shared class state. Twelve of them, including test_non_rotor permanently flipping sched1.job_types['rotors'], test_add_label_to_unique_species_labels appending to a list after asserting its exact initial value, and test_troubleshoot_ess_under_max_attempts reading what test_troubleshoot_ess_max_attempts wrote.

The change

setUpClass → setUp, plus a make_project_directory() helper giving each test a tempfile.mkdtemp under ARC_PATH/Projects released by addCleanup. tearDownClass deleted.

Construction cost was measured before choosing: setUpClass takes 62 ms, so per-test construction is cheap and the simple answer was the right one — no need to split read-only from mutating tests.

Verification

TestScheduler, -n 4 --dist worksteal	22/22 pass (was 0/12)
solo -n 4	6/6
whole file, -n 4 --dist worksteal	5/5
serial test count	45 before, 45 after
serial runtime	4.98 s → 6.70 s (+35 %)
-n 4 runtime	~4.0 s → ~3.9 s (unchanged)

Mutation-verified: restoring the shared fixed-name project directory and its tearDownClass brings the failures back in 3 of 10 runs. Re-sharing the scheduler object alone did not reproduce within 10 runs, because xdist dispatches in collection order so the polluting test usually still precedes its victim on the same worker — the order dependency is instead proven directly by the solo-run result above.

Also fixed: test_save_e_elect cleaned up with a trailing rmtree that leaked its directory on any failure; now addCleanup.

One commit, one file.

Relationship to other PRs

Same defect class as #1008, in a different file — #1008 converts 19 test modules writing scratch into ARC_TESTING_PATH. This one is ARC_PATH/Projects and is independent of it. Fixing this also unblocks CI on #1001, which has been failing on test_initialize_output_dict by luck of the draw rather than on anything in its own diff.

🤖 Generated with Claude Code

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 64.58%. Comparing base (45d73a0) to head (320e542).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1009      +/-   ##
==========================================
- Coverage   64.60%   64.58%   -0.02%     
==========================================
  Files         119      119              
  Lines       39785    39785              
  Branches    10307    10307              
==========================================
- Hits        25703    25696       -7     
- Misses      11105    11110       +5     
- Partials     2977     2979       +2     

Flag	Coverage Δ
functionaltests	64.58% <ø> (-0.02%)	⬇️
unittests	64.58% <ø> (-0.02%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ 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.