[kvlonge]

Why

SQLBuild formats models whose CTE list mixes macro calls and comments, such as:

WITH base AS (SELECT 1 AS order_id),
-- First order stage.
@py.make_ctes("first_orders", "base"),
-- Second order stage.
@py.make_ctes("second_orders", "first_orders"),
@py.other_ctes("third_orders", "second_orders"),
-- Publish order identifiers.
final AS (SELECT order_id FROM third_orders)
SELECT * FROM final

SQLClean 0.6.1 declines it with "native formatter could not restore CTE macro placeholders". A comment before or after a run of placeholders works; a comment between two of them does not.

Changes

• A comment between CTE macro placeholders stays in the SQL laid out without them, so the placeholders after it are now taken out as their own group and put back after it. Before, the whole run went back before the comment and the token check refused the result.
• A comment between a placeholder and its own comma is still declined, as before; the token check catches it.

Verification

• New layout-rule case with that shape on DuckDB: it fails without the fix (formatted false) and passes with it, including the fingerprint and idempotency checks.
• Probed on DuckDB and Snowflake: block comments and a trailing comment after a placeholder's comma also format; every result keeps the authored tokens and comments in order.
• DuckShop's 7 captured format requests give byte-identical responses before and after.
• make verify: fensu 0 faults; cargo 40 + 8 passed; pytest 72 passed.
• SQLQuack stays pinned at 804864d2, the rev SQLBuild's cutover pins.

[cloudflare-workers-and-pages]

Deploying sqlclean with    Cloudflare Pages

Latest commit:	916acb1
Status:	✅  Deploy successful!
Preview URL:	https://77dd9723.sqlclean.pages.dev
Branch Preview URL:	https://fix-cte-macro-comments.sqlclean.pages.dev

View logs