fix: make column mapping order deterministic

[funcpp]

Cherry-picked from #12 (56677a0), one of the contributions in #15. Authorship preserved; the only change on rebase was a conflict caused by the rustfmt pass in #17.

What

resolve collected output nodes into a HashSet<NodeId> and built mappings by iterating it, so construction order followed hash order with a per-process random seed. The sort_by_key afterwards keyed on a HashMap<String, usize> of output names, so duplicate names collided and same-named mappings stayed in whatever order the set produced.

The loop now walks ordered_cols, which already carries projection order. That makes the sort redundant, so it goes too.

SELECT a.id, b.id FROM a JOIN b ON a.id = b.bid

Both mappings target id; which came first varied between runs.

Verification

duplicate_output_names_preserve_projection_order, run 15 times against the pre-fix resolver and the fixed one: 14/15 fail before, 0/15 fail after.

(#12 also carried a three-duplicate variant of the same test. It covered the same path — with the HashSet gone, order follows ordered_cols structurally — so a second commit here drops it. It was measured too: 12/15 fail pre-fix, 0/15 post-fix.)

Checked for behavior change beyond ordering, since removing the HashSet also removes its deduplication and removing the sort could reposition star-expanded columns. Neither shows up — output is identical pre- and post-fix for:

SELECT a, a FROM t                                     ["a", "a"]
SELECT a AS x, a AS x FROM t                           ["x", "x"]
SELECT id, id FROM users                    (catalog)  ["id", "id"]
SELECT *, id FROM users                     (catalog)  ["id", "id", "name"]
SELECT id, * FROM users                     (catalog)  ["id", "id", "name"]
SELECT *, id FROM users                    (no catalog) ["*", "id"]
SELECT users.*, orders.x FROM users JOIN orders (catalog) ["x", "id", "name"]

Each projection item gets its own Output node, so the HashSet was never deduplicating anything; and ordered_cols already placed star nodes where the old sort pushed their expansions.

cargo fmt --all --check                                           exit 0
cargo clippy --workspace --all-targets --all-features -D warnings exit 0
cargo test --workspace --all-features                             exit 0  (91 tests)

Compatibility

No public type or signature change. Results change only for previously nondeterministic cases, where one of the orderings is now always chosen.

🤖 Generated with Claude Code