POK-91: Fix pluck's broken file ops and import-time crash

[saheljalal]

Summary

pluck's characterization suite was failing (9 tests, POK-91), but that was only the visible tip of a tool that was non-functional across the board. This restores pluck to working order and adds tests that actually cover the tool's core path.

Three independent, committed-on-main bugs, all fixed here:

1. Import-time crash (the reported symptom). import tomli_w was unguarded, so when the optional write-side dep was absent the entire module failed to import — taking down every unrelated command and, via the shared test loader, reporting all 9 pluck characterization tests as failing (AttributeError: module '_tool_pluck' has no attribute 'parse_path', etc.). This is exactly what happened in the POK-84 aikit QA venv (base + aikit deps + ruamel.yaml, but no tomli-w). Fixed by importing tomli_w lazily like ruamel.yaml: reading TOML uses stdlib tomllib, and writing TOML without tomli-w now raises a clean PluckError instead of crashing.

2. HANDLERS name collision → every file op broken. The command-dispatch dict HANDLERS (added when pluck was refactored onto scriptkit's sk.dispatch in 1f2b7a06) shadowed the pre-existing format-handler map of the same name. So handler_for() looked a ConfigFormat enum up in the command table and every get/set/copy/merge/diff crashed with an unhandled KeyError on all four formats. No test caught this. Renamed the format map to FORMAT_HANDLERS.

3. TOML write str/bytes mismatch. tomli_w.dumps() returns str, but save()/dump_preview()/the get display path treated it as bytes (path.write_bytes(...) / .decode("utf-8")), so any TOML write or preview raised TypeError/AttributeError. Fixed to write text.

Why the suite didn't catch bugs 2 & 3

The characterization tests pinned only pure helpers + --help — never the dispatch → handler_for → load/save path, which is the whole tool. This PR adds end-to-end get/set round-trip tests across all four formats. They fail on the pre-fix tool (verified by reverting pluck and re-running) and pass after.

Verification

• Full suite: 407 passed (was 9 failed / 390 passed; +8 new round-trip tests).
• Manual: get/set/copy/diff/doctor/formats/--version all exercised across yaml · json · toml · env, including cross-format copy (TOML write).
• Graceful degradation: without tomli-w, --help and the characterization suite still pass, TOML read works, and TOML write fails with a clean ❌ tomli-w is required to write TOML (pip install tomli-w) (no traceback).

Bumps pluck 1.1.0 → 1.1.1 (PATCH — bug fixes, no interface change) with a CHANGELOG entry.

Closes POK-91

🤖 Generated with Claude Code

[saheljalal]

Review — POK-91: pluck import crash & broken file ops

Verdict: Approve, with one should-fix. The diagnosis is right and the three root causes are real, correctly fixed, and I reproduced the before/after behavior in a tomli_w-absent venv (the POK-84 environment).

Verified

• Lazy tomli_w — module now imports clean without the dep (import OK, version 1.1.1, tomli_w is None True). The characterization loader no longer caches a half-initialized module. ✅
• HANDLERS → FORMAT_HANDLERS split (pluck:588, pluck:597) — handler_for now indexes the format map; the command-dispatch HANDLERS at pluck:1106 is distinct. JSON/env set round-trips work end to end. ✅
• str/bytes fix (pluck:546, pluck:549, pluck:707) — tomli_w.dumps() returns str; write_text(..., encoding="utf-8") and .rstrip() on str are correct. ✅
• Write boundary you flagged — no write path assumes the dep at import. pluck set config.toml … without tomli-w fails with a clean PluckError (tomli-w is required to write TOML), rc=1, no traceback. ✅

Should-fix

• pluck:706 (show_value, TOML branch) — the default (non---json) pluck get config.toml key calls toml_writer().dumps(value), so a pure read requires the write dep. In the exact tomli-w-absent venv this issue came from, pluck get config.toml web.port errors instead of printing 8080, while --json works and the file load itself works. This contradicts the PR's own "reading TOML uses stdlib tomllib; only writing needs tomli_w" framing.
  • Why it matters: get is the most common read command; erroring on the default display for a supported read is surprising, and it's the precise scenario POK-91 targets.
  • Suggested fix: when tomli_w is None, degrade the TOML display to a writer-free rendering (reuse the JSON pretty-print, or str(value)) instead of raising — reads should never need the writer.

Nit (optional)

• The new test_pluck_get_reads_value exercises only --json, so the default-display path above stays untested. A default-display assertion (or the fallback fix) would pin the read-without-writer path that this PR is fundamentally about.
• Pre-existing, out of scope: EnvHandler.save (pluck:569–573) has an if/else whose two branches call set_key identically — dead branching, harmless. Not introduced here; noting only.

Nice work closing the test gap that let bugs #2/#3 ship invisibly — the dispatch → handler_for → load/save round-trips are exactly the coverage that was missing. CHANGELOG and the 1.1.0 → 1.1.1 PATCH bump are accurate.

[saheljalal]

Addressed in 164f996 (approved review's should-fix + nit).

Should-fix — show_value TOML branch: the default (non---json) get/diff TOML display no longer calls the writer. It renders as TOML only when the value is a mapping and tomli-w is present, else falls back to JSON — so pluck get config.toml web.port now prints 8080 in a tomli-w-absent venv (reading stays on stdlib tomllib). This also fixed a latent crash the change surfaced: tomli_w.dumps() requires a top-level table, so displaying a bare scalar/list (get config.toml a.scalar) raised AttributeError: 'int' object has no attribute 'items' even with the writer installed — the round-trips missed it by reading via --json.

Nit — untested default display: extended test_pluck_get_reads_value to exercise the default panel path (not just --json), and added test_pluck_get_toml_display_needs_no_writer, which shadows tomli_w as unimportable to pin the read-without-writer path (fails pre-fix, passes after).

EnvHandler dead branch: left as-is per your "pre-existing / out of scope / noting only".

Full suite 408 passed; CI green on 164f996. There were no inline review threads to mark resolved (the review is a single review body), so noting resolution here.