DDL generation grabs column info for the wrong table

[lukasj]

Noticed this in the DDLGenerationExtendTablesTestSuite running sequentially against two different schemas on mysql.

assuming there are two empty DB schemas t1 and t2 on mysql and the test is to end up with a table Countries, with columns iso_code and name where the name column is added by the create-or-extend-tables DDL option

First run on t1 always passes - the particular table gets created with one column, later extended with the new, name, column by the test and all following tests are passing.

Second run on t2 fails with:

Error persisting new country with city, states and zips : jakarta.persistence.PersistenceException: Exception [EclipseLink-4002] (Eclipse Persistence Services - 4.0.0.v202208012021): org.eclipse.persistence.exceptions.DatabaseException
Internal Exception: java.sql.SQLSyntaxErrorException: Unknown column 'name' in 'field list'
Error Code: 1054
Call: INSERT INTO Countries (iso_code, name) VALUES (?, ?)
	bind => [840, USA]
Query: InsertObjectQuery(org.eclipse.persistence.testing.models.jpa.ddlgeneration.Country@705fe9f2)
	at deployment.org.eclipse.persistence.jpa.testapps.ddlgeneration.ear//junit.framework.Assert.fail(Assert.java:57)
	at deployment.org.eclipse.persistence.jpa.testapps.ddlgeneration.ear//junit.framework.TestCase.fail(TestCase.java:223)
	at deployment.org.eclipse.persistence.jpa.testapps.ddlgeneration.ear.org.eclipse.persistence.jpa.testapps.ddlgeneration_ejb.jar//org.eclipse.persistence.testing.tests.jpa.ddlgeneration.DDLGenerationTest.testPrimaryKeyJoinColumns(DDLGenerationTest.java:276)
	at java.base/jdk.internal.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
...

the problem is, that the table in the t2 schema is not extended with the name column.

The cause is that the schemaframework (the part responsible for DDL) does not take into account current schema and catalog when querying the DB for the column names of the particular table, so when it asks for columns of the Countries table, what gets returned is actually info about the Countries table from schema t1 where all columns are already present instead of the same table from schema t2 where the to be created name column is missing, so in the end, the query checking the state of the Countries table in schema t2 fails.