fix(lint): stop two Emacs Lisp rules rewriting quoted symbol lists

[takeokunn]

Survey of the non-Common-Lisp dialect rules — a dimension never audited before. Enumerated by a probe crate iterating the shipped REGISTRY: 358 rules, 76 name a non-CL dialect (58 pure, 18 mixed) = 133 (rule, dialect) sweeps over 9,811 raw → 9,586 SHA-256-deduped files across nine dialects. 0 partial batches; every sweep asserted scanned == list length. 14,579,895 candidate nodes, 19,383 findings.

Rank 1 — elisp-quoted-lambda: Error + Fixable + Destructive, 15/15 false

The predicate was head-only: a list, with a Quote prefix, whose first child is lambda. That matches a symbol list as readily as a quoted function. Demonstrated with the shipped binary against byte-opt.el:

(memq head '(lambda internal-make-closure length cons))
                    ↓ after --fix
(memq head  (lambda internal-make-closure length cons))

A membership test rewritten into a call, automatically, in GNU Emacs's own source. Same shape in bind-key.el, cus-start.el, elint.el, calc-map.el. Two more were ',(lambda …) — menu-bar.el's idiom where the unquote evaluates the lambda so the quote applies to the closure; two were ''(…) / '#'(…).

Fix: require a lambda list ((…) or nil) in second position, and require the prefix to be exactly [Quote]. 15 → 6.

Rank 2 — elisp-obsolete-cl-alias: Error, no context check at all, 25/25 sampled false

150 findings. All binding pairs, lambda lists, or quoted data:

shape	example
binding pair	(dolist (block blocks) …), (let (ll (do t)) …)
lambda list	(defun mail-comma-list-regexp (labels) …)
parameter	(mapcar (lambda (case) …))
quoted data	(memq word '(do doing)), (doctor-type '(do you know Stallman \?))
indent table	cl-indent.el's own spec list

None was a call to a removed macro — unsurprising, since Emacs 31 wouldn't compile if it were.

Fix: require ≥2 arguments; require the first argument to match each macro's real lambda list (List for do/flet, Symbol for block); skip node-local quotes; then ask binding_table().resolve(). 150 → 35 → 14.

A premise in the brief was wrong

I stated that non-CL dialects have an empty binding_table(). Emacs Lisp does not — packages/core/semantics/src/semantics/binding/service/emacs_lisp.rs models named-let, and that check alone removed 21 of 35 residual findings. It holds for LFE (whose rules say so in their own docs), not here.

The residual 14

1 true positive — a genuine unprefixed (case command …) in a chibi-scheme company backend, previously buried in noise. 1 arguable (cl.el's own shim). 12 false, all one class: quote/quasiquote on an ancestor. is_unevaluated_at in lint-form-shape/src/support.rs solves exactly that and #119/#120 established the pattern — but adopting it means copying the two-counter model into a fifth package, which is a separate decision.

Mutation testing earned its place

The second harness first reported 5 survivors — because the new arity guard masked every other guard in the test cases. Rebuilding those controls around real GNU Emacs shapes took it to 7/7 killed (and 6/6 on the first rule). Without it, four guards would have shipped with no coverage.

Controls: 14 asserting every alias still fires in its real call shape; 4 asserting elisp-quoted-lambda still fires on '(lambda (n) n), '(lambda nil 1), '(lambda () 1), &optional forms, and inside a backquote.

All three briefed hypotheses were refuted

hypothesis	result
CL-tuned quote guard on a ,-divergent dialect	not found — the broken rules had no guard, not a wrong one
heads never firing due to case-folding	refuted — 22 rules fired zero times, but all 22 had non-zero candidates, so no false-cleans
firing inside unmodelled Hy quasiquote	refuted — reader_policy.rs:748 documents ~ as deliberately not a prefix, so the effect is suppression; Hy's coverage is understated

Reported, not fixed

• leftover-print-debug — Warning + Fixable, 6,560 findings. Treats Scheme's display and Janet's print as debug leftovers when they are those languages' primary output primitives: (display line out) in an echo server, (print (slurp path)) in an amalgamation tool. --fix deletes them.
• elisp-defcustom-missing-group — fires on 3,668 of 8,476 defcustoms. Correct, but relies on documented file-level defgroup inheritance; belongs in RuleTag::Pedantic.

Verification

Re-running all 133 sweeps against the patched build produced a diff of exactly two lines, both reductions, nothing added.

cargo build / test -p paredit-feature-emacs-lisp (43) / clippy --all-targets -- -D warnings / fmt --check / test -p paredit-cli --lib (334) / test --test cli (3,085 passed) — all exit 0, read unpiped. No golden moved, no pinned count moved.

Known gap: two subagents adjudicating mid-tier rules (elisp-interactive-arity-mismatch, the Scheme/Racket Fixable set, duplicate-test-name, racket-match-unreachable-clause) never returned. Those are surveyed but not adjudicated.