[
https://issues.apache.org/jira/browse/CALCITE-5518?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17688134#comment-17688134
]
Julian Hyde commented on CALCITE-5518:
--------------------------------------
I don't think that 'natural order' is a useful concept here. The fields might
be called 'a' and 'b', and perhaps 'a' maps to ordinal '1' and 'b' maps to
ordinal '0', but all of those things are irrelevant. The behavior of ROLLUP
must not depend on field names or field ordinals.
[~jiajunbernoulli], Your fix makes assumptions about the order of
{{aggregate.getGroupSets()}}. I can see those assumptions are valid by reading
the code of {{Aggregate.Group.isRollup}}, but those assumptions still make the
code fragile and difficult to understand. How about calling
{{Aggregate.Group.getRollup}} instead?
> RelToSql converter generates invalid order of ROLLUP fields
> -----------------------------------------------------------
>
> Key: CALCITE-5518
> URL: https://issues.apache.org/jira/browse/CALCITE-5518
> Project: Calcite
> Issue Type: Bug
> Components: jdbc-adapter
> Affects Versions: 1.32.0
> Reporter: Leonid Chistov
> Assignee: Jiajun Xie
> Priority: Major
> Labels: pull-request-available
> Time Spent: 40m
> Remaining Estimate: 0h
>
> RelToSqlConverter generates wrong code for Aggregate nodes with grouping sets
> classified as `ROLLUP` grouping but with order of rollup not matching order
> of grouping fields.
> This can be demonstrated by the following test, that would fail if added to
> RelToSqlConverterTest class:
> {code:java}
> @Test void testGroupingSetsRollupNonNaturalOrder() {
> final String query = "select \"product_class_id\", \"brand_name\"\n"
> + "from \"product\"\n"
> + "group by GROUPING SETS ((\"product_class_id\", \"brand_name\"),"
> + " (\"brand_name\"), ())\n";
> final String expected = "SELECT \"product_class_id\", \"brand_name\"\n"
> + "FROM \"foodmart\".\"product\"\n"
> + "GROUP BY ROLLUP (\"brand_name\", \"product_class_id\")";
> sql(query)
> .withPostgresql().ok(expected);
> }{code}
> As the result we get the following SQL code:
> {code:java}
> SELECT product_class_id, brand_name
> FROM foodmart.product
> GROUP BY ROLLUP(product_class_id, brand_name){code}
> While the correct code would be:
> {code:java}
> SELECT product_class_id, brand_name
> FROM foodmart.product
> GROUP BY ROLLUP(brand_name, product_class_id){code}
> Source of the bug is that we treat grouping sets \{0, 1}, \{1}, {} as rollup
> grouping, but right after that we generate SQL code as if grouping sets were
> \{0, 1}, \{0}, {}.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
