fix(setup): atomic write for interactive config file

[cmeans-claude-dev]

Summary

Closes #70. Cascades the atomic_write_text helper (introduced in #69) into cli/setup.py:178-183, the last remaining user-visible non-atomic write site in the project. The interactive setup flow previously persisted the user-edited config via path.write_text(...), exposing the same torn-write window the runtime state files had: a Ctrl+C, OOM, or power loss between truncate and final write would leave a zero-byte or half-written <instance_id>.yaml that the next mcp-synology check/serve invocation would fail to parse.

Practical risk is much lower than the runtime-state case (single small-payload write inside an interactive workflow), but the helper was already there ready to use and closing this site means there are no remaining path.write_text calls in the project that persist user/runtime data.

Drops the now-redundant _CONFIG_DIR.mkdir(parents=True, exist_ok=True) since atomic_write_text already creates parent dirs.

Diff at a glance

```python

Before

_CONFIG_DIR.mkdir(parents=True, exist_ok=True)
raw_yaml = yaml.dump(config_dict, default_flow_style=False, sort_keys=False)
header = "# Generated by mcp-synology setup\n"
config_path.write_text(header + raw_yaml, encoding="utf-8")

After

from mcp_synology.core.fs import atomic_write_text

raw_yaml = yaml.dump(config_dict, default_flow_style=False, sort_keys=False)
header = "# Generated by mcp-synology setup\n"
atomic_write_text(config_path, header + raw_yaml)
```

The function-local import follows the convention QA round 1 on #69 explicitly endorsed for cli/version.py:_save_global_state (consistent with the file's existing local-import pattern; setup.py also locally-imports pydantic.ValidationError and mcp_synology.core.config.format_validation_error).

QA

Manual tests

1. • Run uv run pytest tests/core/test_cli_setup.py::TestSetupAtomicConfigWrite -v — should pass.
2. • Run uv run pytest — full suite green at >=95% total coverage.
3. • Run uv run ruff check src/ tests/ scripts/ — clean.
4. • Run uv run ruff format --check src/ tests/ scripts/ — clean.
5. • Run uv run mypy src/ scripts/ — strict-mode clean.
6. • Spot-check that cli/setup.py no longer calls _CONFIG_DIR.mkdir or config_path.write_text. The only remaining write_text calls in the codebase should be in test files. Quick grep: grep -rn "write_text" src/ should return only the new atomic_write_text definition in core/fs.py and no production call sites.

Verification I already ran

Check	Result
uv run pytest	528 passed, 94 deselected, 96.18% total coverage
New test TestSetupAtomicConfigWrite::test_setup_writes_config_atomically_with_no_tmp_sibling	passes; verifies parent-dir auto-creation, no .tmp lingering, header + host content
uv run ruff check src/ tests/ scripts/	clean
uv run ruff format --check src/ tests/ scripts/	clean
uv run mypy src/ scripts/	clean (30 files, strict-mode)

Heads-up

Issue #70 itself was unfortunately filed under cmeans (the maintainer) rather than cmeans-claude-dev[bot] due to a stale-token mistake during PR #69 round 1 — see the ## Heads-up discussion on PR #69 round 2 and the awareness proposal 5b8594c8 for the durable enforcement design that's awaiting maintainer review. The PR itself (this one) is bot-authored as expected.

🤖 Generated with Claude Code

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[cmeans]

LGTM

[cmeans]

QA Round 1 — One substantive finding

PR cleanly cascades #69's atomic_write_text helper into cli/setup.py, the last user-visible non-atomic write site. Issue #70's stated scope is fully covered (helper swap, redundant _CONFIG_DIR.mkdir removed, regression test asserts parent-dir auto-creation + no .tmp lingers). Repo-wide grep -rn write_text src/ confirms the PR body's claim that no production call sites remain outside core/fs.py's own helper internals.

The new test does what the issue's test plan asked for — and a bit more, asserting the resulting YAML has the expected header and contains the entered host. Good.

Verification

Check	Result
uv run pytest	528 passed (+1), 94 deselected, 96.18% coverage
uv run pytest tests/core/test_cli_setup.py::TestSetupAtomicConfigWrite -v	1 passed
uv run ruff check src/ tests/ scripts/	clean
uv run ruff format --check src/ tests/ scripts/	72 files already formatted
uv run mypy src/ scripts/	clean (30 files, strict-mode)
grep -rn write_text src/	only the helper definition + tmp-write inside the helper; 3 production callers all route through atomic_write_text
Required CI on 19caced1	13/13 green

PR-body manual tests 1–6 all checked.

Findings

ID

Finding

F1 (substantive)

Doc drift in src/mcp_synology/core/fs.py:3-5. The module docstring still reads Currently houses atomic_write_text, used by core/state.pyandcli/version.py to persist runtime state without a torn-write window. After this PR cli/setup.py is also a caller — the inventory is now wrong. Fix: either add `cli/setup.py` to the list, or generalize wording (e.g., used by all sites that persist runtime state and user config without a torn-write window). I prefer the latter — once a third site exists, the per-site inventory is liable to drift again on the next cascade. Per the project's doc-drift-is-substantive rule, blocking. Repo-wide grep for the same drift pattern (grep -rn "core/state.py.*cli/version.py" src/ docs/ README.md) confirms this is the only place — no cross-file fix-up needed.

Verdict

QA Failed Round 1 on F1. The fix is a one-line docstring edit in the file you just modified; round 2 should be quick.

[cmeans]

Applying QA Failed as the final act of round 1. F1 is a one-line docstring edit in core/fs.py:3-5 to reflect that cli/setup.py is now also a caller (or generalize the wording so the next cascade doesn't redrift). All 13 CI checks green; PR-body manual tests 1-6 all verified and checked.

[cmeans-claude-dev]

Round 2 — F1 addressed

Generalized the core/fs.py module docstring (commit 5bb8b04) so it no longer enumerates callers by name:

```python

Before

"""Filesystem utilities used across core/.

Currently houses atomic_write_text, used by core/state.py and
cli/version.py to persist runtime state without a torn-write window.
"""

After

"""Filesystem utilities used across core/.

Currently houses atomic_write_text, used by every site that persists
runtime state or user config without a torn-write window.
"""
```

Per the QA preference: drops the per-site inventory entirely so the next cascade (e.g., a fourth caller someday) won't redrift it. Confirmed via grep -rn "core/state.py.*cli/version.py" src/ docs/ README.md that no other doc/source references duplicate the inventory pattern.

Verification

Check	Result
uv run pytest	528 passed, 94 deselected, 96.18% coverage
uv run ruff check src/ tests/ scripts/	clean
uv run ruff format --check src/ tests/ scripts/	72 files already formatted
uv run mypy src/ scripts/	clean

Ready for round 2.

[cmeans]

QA Round 2 — PASS

F1 closed by 5bb8b04 with the generalized-wording option I recommended. New docstring:

"""Filesystem utilities used across `core/`.

Currently houses `atomic_write_text`, used by every site that persists
runtime state or user config without a torn-write window.
"""

Future cascades (more callers added to the helper) won't reintroduce the same drift class. Good shape.

Verification (re-run on 5bb8b04)

Check	Result
uv run pytest	528 passed, 94 deselected, 96.18% coverage
uv run ruff check src/ tests/ scripts/	clean
uv run ruff format --check src/ tests/ scripts/	72 files already formatted
uv run mypy src/ scripts/	clean (30 files, strict-mode)
Required CI on 5bb8b04	13/13 green

Verdict

Ready for QA Signoff. F1 closed; #70 scope still satisfied; no new regressions introduced. Final maintainer call.

[cmeans]

Applying Ready for QA Signoff as the final act of round 2. F1 closed (core/fs.py docstring generalized at 5bb8b04). 528/528 pass, 96.18% coverage, ruff/mypy clean, 13/13 required CI green. Final maintainer call.

[cmeans]

LGTM