DDL generation produces invalid type for ZonedDateTime

[IvanPetkov23]

After upgrading to Ebean 16.0.0+, DDL generation for ZonedDateTime fields annotated with @WhenCreated / @WhenUpdated produces timestamp(255) when running against H2.

Since this is not a valid H2 type, schema creation fails in tests.
Is this expected in 16.x, or should this be reported as a bug?

While looking into this further, I noticed that DdlGenerator does not seem to support running only initSql / seedSql without full DDL generation. When generate=false and only initSql is used, a NullPointerException occurs, which suggests these steps are currently coupled to DDL generation.

#https://github.com/ebean-orm/ebean/blob/master/ebean-ddl-generator/src/main/java/io/ebeaninternal/dbmigration/DdlGenerator.java#L126-L129

Is this expected, or should running initSql / seedSql independently be supported?

[rbygrave]

ZonedDateTime fields annotated with @WhenCreated / @WhenUpdated produces timestamp(255) when running against H2.

You should create a failing test.

Noting that this doesn't reproduce in master, so you should create a failing test first to confirm you can reproduce the issue.

  @WhenCreated
  ZonedDateTime zonedDateTime;
  @WhenModified
  ZonedDateTime zonedDateTime2;

  zoned_date_time               timestamp not null,
  zoned_date_time2              timestamp not null,

When generate=false and only initSql is used, a NullPointerException occurs, which suggests these steps are currently coupled to DDL generation.

These are expected to be used together for testing purposes yes. You might want to outline the use case and include the NPE to see what should happen / if we need to change or fix something here.

[IvanPetkov23]

Here is a minimal reproducer demonstrating the failing test:
ebean-failing-test

You might want to outline the use case and include the NPE to see what should happen / if we need to change or fix something here.

No specific use case — this came up while debugging a DDL generator issue and trying to initialize the database via initSql.

[IvanPetkov23]

Hey @rbygrave,

Whenever you get a chance, could you confirm whether this is something that should be tracked as an issue? I’d be glad to open one if that’s the preferred workflow.

[rbygrave]

Ok, I have reproduced. So the issue is a regression related to https://github.com/ebean-orm/ebean/pull/3619/changes ... which was a feature/change to allow @Column with no length to be mapped to a Postgres varchar with no length. That meant that @Column(255) for Postgres would map to varchar(255).

Background: JPA @Column has a magic default length of 255 ... so that was a problem for this feature.

So as part of that change, Ebean's JPA API needed to set that default length on column from 255 to 0.

However, that change has introduced this regression with H2. So now we need to figure out how to resolve that and keep the Postgres feature.

[rbygrave]

Workaround

The workaround you can use is to remove the dependency on jakarta.persistence:jakarta.persistence-api:3.2.0. Remove that, means it will use the ebean supplied dependency and will be all good.

[rbygrave]

Alternative fix

For completeness, noting there is an alternative fix with PR #3741 ... that detects the version of @Column and hence the default column length of 255.

This alternative approach wasn't chosen for the fix, and we will close that PR but wanted to just include the link in case we want to refer back to it at some future point.