[kvlonge]

D150 diagnostics, plus D171's parse and check fixes (D177): SQL that Snowflake rejects now fails sqlquack check instead of reaching translation, and parse failures are worded for readers.

Stacking. This branch sits on the corpus series (wip/formatter-corpus-series 4070812), which lands first. Until it does, this PR's diff includes the series' commits. Each of the five commits below stands on its own and is meant to be fast-forwarded as is.

Commits

1. D150: check unknown relations and functions, and word parse failures.
   • check --schema: a relation the schema doesn't declare is SQ0004. A call to a function that is neither a documented Snowflake function nor a schema function is the new SQ0205 unknown_function (Snowflake 2140). Without a schema, names stay unchecked.
   • An unresolved double-quoted name gets the help line "use single quotes for a string literal: '…'".
   • SQ0001 now says what was expected and what was found. It points an unclosed ( at where it opens, points end-of-input failures past the last token, and adds help for dbt/Jinja templates.
2. == is a parse error in Snowflake SQL (Snowflake 2139), with the help line "Snowflake compares with a single =". Other dialects keep ==.
3. IN 'x', >> and << are parse errors (Snowflake 1003). Scripting's FOR … IN is unchanged. &, | and ^ still parse, because Snowflake parses them and fails at compile (2139 and 2, import 163).
4. An aggregate FILTER (WHERE …) clause is a parse error (Snowflake 1003, at the (). The FILTER function over an array is unchanged.
5. Three schema-free checks, which also stop translation in the playground:

Code	Name	Snowflake errno	What it checks
SQ0206	duplicate_alias	2027	Two FROM items or joins share one alias, compared as Snowflake stores names.
SQ0207	too_many_arguments	939	A call passes more arguments than every catalogue signature allows. snowflake_function_arities.txt lists the 772 functions whose call grammars all have a fixed number of plain arguments; a test derives it from the catalogue. Calls with named arguments, and calls with too few arguments, are not checked: receipts show two documented minimums (ARRAYS_ZIP, FORMAT_STRING_TEMPLATE) aren't enforced.
SQ0208	pivot_consumed_column	904	The query reads a column that a lone PIVOT aggregates or pivots on, outside SELECT-list aliases and nested queries.

Evidence that nothing valid is refused

• Parser changes (==, IN, >>, <<, FILTER): the parse verdict changes for none of the 56,748 receipted statements (through import 161). Receipts exercise these paths: 3,981 statements have a bare IN, mostly scripting FOR … IN, and 11 have ->>.
• SQ0206, SQ0207, SQ0208: none fires on any of the 24,472 parsed receipted statements, or on the 4,243 statements of the translation cases (3,292 of their calls are covered by the arity table). Among them are 98 receipted statements and 6 cases that use PIVOT.
• SQ0207 confirmed against Snowflake (import 163, wip/syntax-evidence 462b572): 23 calls, each passing one argument more than the table's limit, all fail with 939/22023 "too many arguments". They cover LENGTH, SUBSTR, DATEADD, IFF, TRIM, FLOOR, TRUNC, REPLACE, REGEXP_LIKE, TIME_SLICE and 13 others. The control ZIPF ( 1 , 2 , RANDOM ( ) ) compiles; ZIPF is excluded because its page grammar has an empty slot.
• SQ0205: fires only when a schema is given. The function list is kept in sync with the catalogue by a test.

Everyday corpus

7 of the 8 queries that Snowflake rejects but SQLQuack used to pass now fail at parse or check. Their expect lines are re-derived, and their entries are removed from expected_rejections_not_caught.

Query	Before	After
WHERE status == 'shipped'	translated	parse SQ0001
WHERE status IN 'shipped'	refused SQ0002	parse SQ0001
WHERE total >> 10	refused SQ0002	parse SQ0001
COUNT(*) FILTER (WHERE …)	refused SQ0002	parse SQ0001
orders o, orders o	refused SQ0004	check SQ0206
LENGTH(order_id, 2)	refused SQ0002	check SQ0207
PIVOT, then total	refused SQ0003	check SQ0208
FROM order, FROM orderz	refused SQ0004	check SQ0004
NOSUCHFUNCTION(order_id)	refused SQ0002	check SQ0205

WHERE ordered_at > 5 (1038) stays a translation refusal: catching it at check needs typed checking over the schema. LIMIT read as a table alias waits on Snowflake receipts and will follow in its own change.

Checks run

• The core, semantics, syntax, format and CLI test suites.
• clippy with -D warnings.
• fensu: 0 faults.
• The playground tests: 8/8.
• The public repository hygiene scan: exit 0.
• website/scripts/generate_diagnostics.py, which regenerated reference.mdx.

[kvlonge]

Closing: SQLQuack lands through local-gate trains, not PR merges (D94), so open PRs only add clutter (D628/D630). The branch is kept. Any work here that isn't on main yet is tracked by the train owner and lands through a train.