parser: disambiguate bare-token compound patterns in defn (eigentrust pitfall #7)

[kumavis]

Fixes pitfall #7 from docs/tracking/2026-04-23_eigentrust_pitfalls.md.

Summary

defn sum-rows
  | nil            -> nil
  | cons r nil     -> r
  | cons r rest    -> [add-vec r [sum-rows rest]]

failed with Unbound variable sum-rows::1 — the parser treated cons r rest as three separate arguments instead of one compound pattern.

Approach (a) — auto-pack constructor patterns

In parse-defn-clause, when the leading token after | names a known constructor whose field count matches the remaining tokens, pack them into one compound pattern. Falls back to the existing N-arg interpretation when the leading token is not a constructor (e.g., defn add | x y -> [+ x y] still works).

Local change in one branch of the cond, ~24 lines, no other files touched.

Why (a) over (b) "raise a clear error": defn is the primary dispatch mechanism per prologos-syntax.md, and ML/Haskell users naturally read cons r nil as one compound. Detecting via lookup-ctor is local and preserves multi-arg defns unchanged.

Files changed

• racket/prologos/parser.rkt (+24)
• racket/prologos/tests/test-defn-multiarg-patterns.rkt (new, 315 lines, 11 cases)

Test plan

• 11/11 new tests pass
• 43/43 existing tests in test-pattern-defn-01 / test-pattern-defn-02 / test-multi-body-defn pass
• Full affected suite: 4646 tests in 255 files; 1 unrelated pre-existing failure (stale tracking entry for non-existent test-constraint-retry-propagator.rkt)
• CI fix included (benchmarks/micro/info.rkt skip; redundant once PR Migrate bench-bsp-le-track2.rkt to current ATMS API (CI unblock) #10's actual migration lands first)

Latent bug surfaced (NOT introduced) — worth tracking

The fix exposed a latent compile-match-tree bug: variable patterns named in a sub-position get bound via let v := __cons_1, referring to a param that's been destructured by a later dispatch column. This corrupts recursive bodies like cons r rest -> [+ r [recurse rest]].

Reproducible on main with the bracketed [[cons r rest]] form (same internal AST). The tests cover the slices unaffected (empty/singleton inputs, wildcards, all-var multi-arg). The commit message documents the latent issue for future tracking — it should be filed as a separate bug.

Commits

1. 1595905 — primary fix in parser.rkt + test file
2. dd26ad4 — CI fix (skip stale bench)

https://claude.ai/code/session_01MbncYJnrvjzhbVWw4xGi5x

---

Generated by Claude Code

[hierophantos]

LGTM. Surgical scope, uses existing lookup-ctor / ctor-meta-field-types infrastructure (no new APIs), N-arg fallback preserved.

Two non-blocking observations:

1. The auto-pack rule means a user who picks a parameter name matching a constructor — defn foo | cons x y -> ... with cons as a variable — would now get the compound interpretation. ML/Haskell-aligned default is right, but worth a release note when a batch of these pitfall fixes lands.
2. Filed the latent compile-match-tree variable-binding bug as #18 with your reproducer + diagnosis attributed. That tracks the recursive eigentrust case independently — your parser fix here can land without it.

Approving — ready to merge once you rebase to drop the skip commit.

[kumavis]

defn sum-rows
  | nil            -> nil
  | cons r nil     -> r
  | cons r rest    -> [add-vec r [sum-rows rest]]

expected to see an argument body following the function name.