[sineline]

Follow-up to #5722, related to #5721.

Snowflake rejects ALTER TABLE for Iceberg tables and requires ALTER ICEBERG TABLE instead. SnowflakeEngineAdapter._create_column_comments builds its statement with a hardcoded TABLE kind, so registering column comments on an Iceberg table with a post-creation command emits:

ALTER TABLE "t" ALTER COLUMN "a" COMMENT '...'

The call is wrapped in try/except, so the plan does not fail: the comments are silently dropped and a warning misattributes the failure to limited permissions.

This path is only taken when the comments cannot be inlined into the CTAS schema definition, i.e. when the column types are not fully known. When the types are known, the comments are part of the CREATE ICEBERG TABLE statement and are registered correctly.

Changes

• EngineAdapter._create_column_comments accepts table_format, mirroring the _create_table convention, and the base adapter passes it from the two table creation paths (create_table and _create_table_from_source_queries).

• SnowflakeEngineAdapter._create_column_comments derives the Iceberg-specific kind from it, only when the kind is TABLE, so views are unaffected:

  ALTER ICEBERG TABLE "t" ALTER COLUMN "a" COMMENT '...'

• The MySQL and BigQuery overrides gain the same parameter so their signatures stay compatible with the base class; they do not use it.

Reference: https://docs.snowflake.com/en/sql-reference/sql/alter-iceberg-table#syntax ({ ALTER | MODIFY } [ COLUMN ] <col_name> COMMENT '<string>', comma-separated for multiple columns).

Tests

• test_column_comments_iceberg: a direct call emits ALTER ICEBERG TABLE ... ALTER COLUMN ... COMMENT ... for multiple columns.
• test_ctas_column_comments_iceberg: an end-to-end ctas with unknown column types emits CREATE ICEBERG TABLE ... AS SELECT ... followed by the ALTER ICEBERG TABLE comment statement.

The generated DDL is verified at unit level against the documented syntax; it has not been validated against a live Snowflake account.

[StuffbyYuki]

@sineline Thanks for this PR! The core fix is good against Snowflake's documented syntax, and regression risk is low. But here's a few suggestions:

1. Dynamic Iceberg tables

The new _create_column_comments call gets table_format but not table_kind, so it defaults to "TABLE". As a result, dynamic Iceberg tables now get ALTER ICEBERG TABLE instead of ALTER TABLE. That syntax isn't documented for dynamic tables, and the try/except would silently drop the comments. A one-line change keeps the fix scoped to plain Iceberg tables:

table_format=None if table_kind else kwargs.get("table_format"),

Could you add a create_managed_table(..., table_format="iceberg", column_descriptions=...) test that asserts ALTER TABLE is emitted?

2. A quick run against a real Snowflake Iceberg table would confirm the SQL beyond the unit tests. (You can spin up a free snowflake instance)