[hibernate-commits] [hibernate/hibernate-orm] 21e791: HHH-9301 - Support select variable refs in group b...

[Jan-Willem Gmelig Meyling]

  Branch: refs/heads/master
  Home:   https://github.com/hibernate/hibernate-orm
  Commit: 21e79125e601160101f1f27e1aea026ede9c36a9
      https://github.com/hibernate/hibernate-orm/commit/21e79125e601160101f1f27e1aea026ede9c36a9
  Author: Jan-Willem Gmelig Meyling <jan-willem at youngmediaexperts.nl>
  Date:   2019-11-25 (Mon, 25 Nov 2019)

  Changed paths:
    M hibernate-core/src/main/antlr/hql-sql.g
    M hibernate-core/src/main/java/org/hibernate/dialect/Dialect.java
    M hibernate-core/src/main/java/org/hibernate/dialect/H2Dialect.java
    M hibernate-core/src/main/java/org/hibernate/dialect/MySQLDialect.java
    M hibernate-core/src/main/java/org/hibernate/dialect/PostgreSQL81Dialect.java
    M hibernate-core/src/main/java/org/hibernate/hql/internal/ast/HqlSqlWalker.java
    A hibernate-core/src/test/java/org/hibernate/query/GroupByAliasTest.java
    M hibernate-testing/src/main/java/org/hibernate/testing/DialectChecks.java

  Log Message:
  -----------
  HHH-9301 - Support select variable refs in group by for DBMS dialects that support it

While not strictly compliant with the SQL specification, the MySQL, PostgreSQL and H2 support the use of select aliases in the GROUP BY clause. An obvious benefit is that produced SQL queries will be easier to read, because complex select expressions from aggregrate queries will have to be included in the group by clause as well. These can now simply reference the aliases of the respective columns for the tuple element. However, there is also a functional difference. For function invocations that have parameterized arguments, the query optimizer can't guarantee the that the function result between the selection projection and grouping process are equal. This results in an error because a value is projected which is not grouped by. An example where this for example becomes relevant, is when parameterizing TimeScaleDB's [`time_bucket_gapfill()`](https://docs.timescale.com/latest/api#time_bucket_gapfill-examples) function.

For example, using `time_bucket_gapfill()` one might want to produce a criteria query that produces the following SQL:

```sql
SELECT
    time_bucket_gapfill(?, time, ?, ?) AS ts,
    avg(tg) as tg
FROM iaqmeasurement
GROUP BY ts
ORDER BY ts;
```

When the alias is not used as grouping value, the query will yield an error:

```sql
SELECT
    time_bucket_gapfill(?, time, ?, ?) AS ts,
    avg(tg) as tg
FROM iaqmeasurement
GROUP BY time_bucket_gapfill(?, time, ?, ?)
ORDER BY ts;
```

Of course the parameter values can just be stored in a CTE as well, but I think we should consider support for variable refs in group by for DBMS dialects that support it. This pull request implements the feature equivalently to how its currently done for order by clauses, and only enables the feature for known supported dialects (H2, PostgreSQL and MySQL - based on [https://stackoverflow.com/a/3841804/2104280](https://stackoverflow.com/a/3841804/2104280)).

Jira: https://hibernate.atlassian.net/browse/HHH-9301

Co-authored-by: Sayra Ranjha <S.S.Ranjha at student.tudelft.nl>