Hi All,
I’ve put the minimal requirements and golden SQL tests in a tiny repo:
*
Repo/branch: DotKBoy-web/acbp-calcite-tests, branch calcite-swap
*
How to run: mvn -DskipTests=false test (4 hand-printer tests + 4 Calcite-backed
tests)
*
The Calcite path uses RelBuilder → RelToSqlConverter → SqlDialect; outputs
match the goldens modulo harmless formatting (identifier quotes, col AS col,
and optional WHEN (...) parens). Tests normalize only those diffs.
I’d love guidance on a few specific knobs / best practices so I can drop most
normalization and keep goldens stable:
1.
Minimal quoting
How do I configure RelToSqlConverter / SqlPrettyWriter / dialect to avoid
quote-all and emit unquoted identifiers where legal, only quoting when required?
2.
Suppress identity aliases
Can I disable redundant aliases like SELECT code AS code (also inside
subqueries) via a public flag/hook? If not, where’s the preferred extension
point to add one?
3.
WHEN parentheses policy
What’s the recommended way to get WHEN a AND b THEN ... instead of WHEN (a AND
b) THEN ... when precedence makes parens unnecessary?
4.
Time-window idioms per dialect
What’s the canonical Calcite construction to unparse exactly these forms?
*
PostgreSQL: now() - interval 'N days'
*
BigQuery: TIMESTAMP_SUB(CURRENT_TIMESTAMP(), INTERVAL N DAY)
*
ClickHouse: now() - INTERVAL N DAY
5.
Dialect settings for BigQuery & ClickHouse
Are BigQuerySqlDialect and ClickHouseSqlDialect the right choices for these
idioms, and are there recommended writer/converter configs to match the styles
above (esp. quoting & aliasing)? Any known limitations I should lock down with
tests?
6.
Golden stability
Which writer/converter options are most stable across Calcite versions? If
absolute stability isn’t guaranteed, what normalization policy aligns best with
Calcite’s formatting choices?
7.
If no public knobs exist
Would a small PR adding opt-in flags (e.g., minimal-quoting,
suppress-self-alias, minimal-parens) be welcome? If so, should they live on
SqlWriterConfig, RelToSqlConverter.Config, or per-dialect overrides?
Pointers to examples or tests in Calcite that demonstrate these behaviors would
be perfect. I’m happy to turn any edge we hit into a focused failing test (one
per dialect) or contribute the needed hooks.
Thanks!
—Muteb (DotK)
