Fix InPlaceRestartMember.setup() work_dir isolation

[Copilot]

InPlaceRestartMember.setup() must never reassign self.work_dir—if it did, compass/setup.py would write step.pickle into the spinup run directory containing config_filename = sgh_restart_ensemble.cfg, causing compass run invoked from that directory to crash with FileNotFoundError: Config file does not exist: .../spinup_ensemble/run002/sgh_restart_ensemble.cfg.

Changes

• restart_member.py — setup() docstring: Rewritten to explicitly document that the method operates exclusively on self.spinup_run_dir (set in __init__) and never touches self.work_dir, which compass owns as the step directory. Removed a verbose inline comment made redundant by the updated docstring.
• Verified (no change needed):
  • __init__ has no self.work_dir override
  • ensemble_manager.py uses getattr(runStep, 'spinup_run_dir', runStep.work_dir) for os.chdir()

Checklist

• User's Guide has been updated
• Developer's Guide has been updated
• API documentation in the Developer's Guide (api.rst) has any new or modified class, method and/or functions listed
• Documentation has been built locally and changes look as expected
• The E3SM-Project submodule has been updated with relevant E3SM changes
• The MALI-Dev submodule has been updated with relevant MALI changes
• Document (in a comment titled Testing in this PR) any testing that was used to verify the changes
• New tests have been added to a test suite

Original prompt

Problem

compass/setup.py sets step.work_dir to the correct compass step directory before calling step.setup(), then writes step.pickle to step.work_dir after setup() returns. However, InPlaceRestartMember.setup() reassigns self.work_dir to the spinup run directory at the very start:

# restart_member.py lines 95-96 — THIS IS THE BUG
self.work_dir = os.path.join(
    self.spinup_work_dir, f'run{self.run_num:03}')

So by the time setup.py writes step.pickle, step.work_dir points to the spinup run dir, and the pickle is written there — containing an InPlaceRestartMember with config_filename = sgh_restart_ensemble.cfg. When the spinup job_script.sh runs compass run from that directory, it loads this pickle and crashes:

FileNotFoundError: Config file does not exist:
  /pscratch/.../spinup_ensemble/run002/sgh_restart_ensemble.cfg

Root cause

PR #12 removed the self.work_dir override from __init__ but missed the identical override at the top of setup(). That override in setup() is what setup.py sees when it writes step.pickle after setup() returns.

Fix

In compass/landice/tests/ensemble_generator/sgh_restart_ensemble/restart_member.py:

1. Delete lines 95–96 (the self.work_dir = os.path.join(...) assignment at the start of setup()).

2. Replace all remaining uses of self.work_dir and run_dir in setup() with self.spinup_run_dir, which was added by PR Fix InPlaceRestartMember overwriting spinup job_script.sh #12 in __init__ and already holds the correct path to the original spinup run directory.

The resulting setup() should look like:

def setup(self):
    """
    Prepare the original run directory for an in-place restart.

    Uses ``self.spinup_run_dir`` (set in ``__init__``) to operate on the
    original spinup run directory without touching ``self.work_dir``, which
    compass manages as the normal step directory.

    This method:

    1. Verifies the spinup run directory and namelist.landice exist.
    2. Sets config_do_restart = .true. in namelist.landice.
    3. Creates a restart_attempt_N/ tracking directory.
    """
    run_dir = self.spinup_run_dir

    if not os.path.exists(run_dir):
        raise RuntimeError(
            f"Original run directory not found: {run_dir}")

    namelist_path = os.path.join(run_dir, 'namelist.landice')

    if not os.path.exists(namelist_path):
        raise RuntimeError(
            f"namelist.landice not found in {run_dir}")

    print(f'Setting config_do_restart = .true. in {namelist_path}')
    _set_restart_in_namelist(namelist_path)

    existing_nums = [
        int(d[len('restart_attempt_'):])
        for d in os.listdir(run_dir)
        if d.startswith('restart_attempt_') and
        d[len('restart_attempt_'):].isdigit()
    ]
    attempt_num = max(existing_nums, default=0) + 1
    attempt_dir = os.path.join(run_dir, f'restart_attempt_{attempt_num}')
    os.makedirs(attempt_dir, exist_ok=True)
    print(f'Tracking restart attempt {attempt_num} in {attempt_dir}')

Also verify that __init__ does NOT contain any self.work_dir override (it should not after PR #12 — just confirm and leave it as-is if correct).

Also verify that ensemble_manager.py uses getattr(runStep, 'spinup_run_dir', runStep.work_dir) for the os.chdir() call (added by PR #12 — just confirm and leave as-is if correct).

Current file state

• compass/landice/tests/ensemble_generator/sgh_restart_ensemble/restart_member.py BlobSha: f13e64c3a82e5f3cc122eed7b8175ab0a556abfb

Constraints

• PEP 8 formatting
• No fallback= in any configparser calls
• No timeout= in any subprocess calls

The following is the prior conversation context from the user's chat exploration (may be truncated):

User: can you relink the agent session? It's blank

Assistant: The latest PR is #12:

data:
- url: "https://github.com/alexolinhager/compass/pull/12"
  repository: "alexolinhager/compass"
  state: "open"
  draft: true
  title: "[WIP] Fix job script overwrite in setup process"
  number: 12
  created_at: "2026-04-01"
  merged_at: ""
  labels: []
  author: "Copilot"

It's still in WIP/draft — the agent is actively working on it. The checklist shows the two tasks (fix restart_member.py and fix ensemble_manager.py) are pending. You can track progress directly at #12.

User: These fixes are still not getting to the root of the problem:
Traceback (most recent call last):
File "/global/cfs/cdirs/fanssie/users/ahager/mambaforge/envs/dev_compass_1.9.0-alpha.2/bin/compass", line 7, in
sys.exit(main())
~~~~^^
File "/global/cfs/cdirs/fanssie/users/ahager/MALIcode/compass/compass/main.py", line 63, in main
commandsargs.command
~~~~~~~~~~~~~~~~~~~~~~^^
File "/global/cfs/cdirs/fanssie/users/ahager/MALIcode/compass/compass/run/serial.py", line 206, in main
run_single_step(args.step_is_subprocess)
~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^...

This pull request was created from Copilot chat.