[cmj0121]

Closes #295. Last task of #285.

What changes

== and != over two values of one unnamed form are built: a list and an array compare in order, a tuple
part by part, and a map or a set whatever order they were filled in. E9057 retires.

The rule is the one docs/core/types.md already had — an unnamed form has Eq exactly when every part has
it — so one predicate (c_eq_legal) answers the operator and the T: Eq bound alike.

• A part is compared by the == a program would write between two of them. The two parts are bound as
  two names and the operator is handed back to c_expr, so a str, a declared type's eq (with the copies
  that call owes) and a form inside a form are each decided where they already were.
• A map's keys are found, not compared: by the hash and equality the map was built with.
• != is the negation. An unnamed form wrote no ne to dispatch to.
• What is refused is the part, named beside the form, under the code that part already had:
  E4020 ==on a list[N], which compares its N: a N has noEq``, E4038 for a carrier, `E4034` for an
  identity.

Decided here, and worth a second look

• Tuples and arrays are in. two containers compare with == and != #295 names list, map and set; E9057 also covered tuples, the types
  chapter gives a tuple Eq by the same rule, and the code cannot retire with a tuple still under it.
• A literal does not take the other operand's element type. bs == [1, 2] over a list[byte] is
  E3046, and xs == [] is E3034. Both were already what the checker said; the collections chapter now
  says so.

Not in this PR

• x in xss over a list of containers is still E9061, and over a list of a declared type with an Eq.
• assert xs == ys over two lists is E4011: the comparison builds, and the message has to render both
  sides, which str over a list does not. A map, a set, a tuple and a declared type are claimed directly.
  The package chapter says so and gives assert (xs == ys).
• An ordering over two tuples is E3044, while the types chapter gives a tuple Ord by its parts. Not
  touched, not filed.

Measured

Each gate run alone, its own status read: build fixpoint lint fmt-self error-codes-check
seed-gaps refuse reject mem-check oracle corpus sanitize-corpus mem-peak examples
docs-mirror docs-links docs-zerg docs-refused docs-repeat marker-codes chapter-codes
deviation-check refusal-cites method-gaps gates treesitter dead-code stdlib-test
compiler-test check-equal reject-fuzz fmt-corpus behaviour conformance productions suites
desugar — all OK. refuse-check, reject-check and mem-check also under /bin/bash 3.2.

• mem-check case container_equality: RED with the left operand's give-back removed
  (5070 per round unfreed), green with it.
• lsp: FAIL through make lsp on this machine, OK through a copy of the script that keeps its fixtures
  in an empty directory. The server lists the directory above the buffer, zrt_listdir sorts
  quadratically, and this machine's temp directory holds forty thousand entries. Not this change, and not
  filed.
• The whole board (make test) was not run.

The seed compares no container, so the corpus case is an oracle skip.

🤖 Generated with Claude Code

https://claude.ai/code/session_01N8jd1oK2orM6diHxhvWhC7