[kvlonge]

Launch blocker D327 (from D320): comments must never change whether a statement converts.

The bug

SELECT /* c */ 1 AS v and SELECT -- note\n 1 AS v refused the whole statement with SQ0002, "no rule handles post_select_comments set on node:select". That was 602 of the playground lane's 3,263 generated statements. Snowflake compiles both and returns column V (A's receipts, wip/d320-playground-basics 6111431), so the refusal was ours alone.

The fix

Both changes are in conversion/_helpers/support/tree.rs, which already decides which tree fields are cosmetic:

1. post_select_comments joins COSMETIC_FIELDS, beside trailing_comments, pre_alias_comments and the others. The comment is dropped from the output, the same as comments everywhere else.

2. The sweep turned up a second case. A comment straight after an expression wraps it in an annotated node that holds only this and trailing_comments. A rule that needs a literal then sees the wrapper: ROUND(amount, 1 /* c */) refused with SQ0003 while ROUND(amount, 1) converted. When the tree is built, a wrapper that holds nothing but comments is now replaced by the node it wraps.

3. Round 1 review (auditor record 09453d8) found that unwrapping also renamed unaliased items: SELECT 'x' /* c */ FROM orders took the name 'X'. Main had the same gap, since it named it 'X' too; no receipt records whether Snowflake's name includes the comment. A select item's own wrapper now stays (unwrap_field), and the projection rule refuses it with SQ0009 unless it wraps a column, star or alias, as the column-name rule's conditions already state. SELECT 'x' /* c */ AS v and SELECT amount /* c */ still convert.

Sweep

38 comment positions were converted over a schema (orders, customers). With only change 1, 1 of the first 32 still refused (the comment after ROUND's literal scale); with both changes, 0 of 38 refuse. I didn't run the sweep on unchanged main; the after-SELECT positions are the D320 report's. The positions:

• after SELECT (block and line), before or after DISTINCT, between select items, after an alias or AS
• after FROM, after the table, before WHERE (line), after WHERE, inside a comparison
• after GROUP BY / ORDER BY / LIMIT / HAVING / QUALIFY, after a CTE name, after SELECT inside a CTE and inside a subquery
• before a CTE's, a subquery's and a function's closing paren, inside function calls, after JOIN and ON, in CASE and in a window
• after UNION ALL, after the second SELECT of a UNION, trailing at the end (block and line)
• after literals: a select item, LIMIT, an IN list, a string argument, a comparison, a date part

Tests

test_analyze gains five rows: a block and a line comment after SELECT, comments after SELECT in a CTE and a subquery, a comment after a literal argument, and an unaliased literal carrying a comment (SQ0009, which fails with query.rs reverted to main). With tree.rs reverted to main, the first new row fails; with the fix, all pass.

Gate

D94 gate on exact head 9c008e0 (58731ec + merge of main fc15d51, clean, no conflicts), every .rs touched before the build:

• make check-merge (check-ci + test-rust): exit 0. fmt, clippy, ruff, ty and fensu clean; 24/24 test binaries passed.
• sqlquack verify --require-verified --strict-errors --python .venv/bin/python: 12,033 cases, 7,997 passed, 1,035 diverged as expected, 2,964 refused, 37 typed, 0 pending, 0 unproven, 0 failed; 226 of 226 equivalences and rules verified.
• syntax-coverage results … --against main fc15d51's results-baseline.json: 0 regressions, 0 other changes; no baseline re-record needed.
• Hygiene scan clean; no agent trailers.

Auditor: approved at 7392627 (record 6a306de), extended to 58731ec (record 976fb43). GitHub Actions is blocked by billing; the local D94 gate counts.

[kvlonge]

D94 gate on exact head 9c008e0 (58731ec + merge of main fc15d51, clean, no conflicts), every .rs touched before the build:

• make check-merge (check-ci + test-rust): exit 0. fmt, clippy, ruff, ty and fensu clean; 24/24 test binaries passed.
• sqlquack verify --require-verified --strict-errors --python .venv/bin/python: 12,033 cases, 7,997 passed, 1,035 diverged as expected, 2,964 refused, 37 typed, 0 pending, 0 unproven, 0 failed; 226 of 226 equivalences and rules verified.
• syntax-coverage results … --against main fc15d51's results-baseline.json: 0 regressions, 0 other changes; no baseline re-record needed.
• Hygiene scan clean; no agent trailers.