[General] Datanucleus is incompatible with Cloud Spanner JDBC Driver >= 2.8.0 + GoogleSQL

[Noremac201]

Environment details

1. Specify the API at the beginning of the title. For example, "BigQuery: ..."). General, Core, and Other are also allowed as types
2. OS type and version: Linux
3. Java version: 1.8
4. version(s): 2.20.1

Steps to reproduce

1. I'm not at liberty to share the code, however, when attempting to use Datanucleus with CloudSpanner JDBC driver newer than 2.8.0 Datanucleus will be unable to get the type info from the JDBC connection string.

Code example

When Datanucleus attempts to load a JDBC connection, it will get the Type info from the connection, however, this type info is then immediately cast to a short in Datanucleus (1). However, with Google SQL and support for JSON it will crash upon trying to convert 100011 to a short, which is coming from the JSON vendor type number (2).

1. https://github.com/datanucleus/datanucleus-rdbms/blob/master/src/main/java/org/datanucleus/store/rdbms/schema/SQLTypeInfo.java#L137
2. https://github.com/googleapis/java-spanner-jdbc/blob/v2.15.5/src/main/java/com/google/cloud/spanner/jdbc/JsonType.java#L33

Stack trace

Caused by: com.google.cloud.spanner.jdbc.JdbcSqlExceptionFactory$JdbcSqlExceptionImpl: OUT_OF_RANGE: Value out of range for short: 100011
    at com.google.cloud.spanner.jdbc.JdbcSqlExceptionFactory.of(JdbcSqlExceptionFactory.java:247) ~[google-cloud-spanner-jdbc-2.20.1-single-jar-with-dependencies.jar:2.20.1]
    at com.google.cloud.spanner.jdbc.AbstractJdbcWrapper.checkedCastToShort(AbstractJdbcWrapper.java:237) ~[google-cloud-spanner-jdbc-2.20.1-single-jar-with-dependencies.jar:2.20.1]
    at com.google.cloud.spanner.jdbc.JdbcResultSet.getShort(JdbcResultSet.java:330) ~[google-cloud-spanner-jdbc-2.20.1-single-jar-with-dependencies.jar:2.20.1]
    at org.datanucleus.store.rdbms.schema.SQLTypeInfo.<init>(SQLTypeInfo.java:133) ~[datanucleus-rdbms-4.1.1.jar:?]
    ... 59 more

The question is, can we either bring down the 100,000 addition to 32,000 ? Would that be a breaking change? This would ensure compatibility to those jdbc loaders who assume JDBC types are very small always.

[olavloite]

@Noremac201 Thanks for the detailed and clear report.

The question is, can we either bring down the 100,000 addition to 32,000 ? Would that be a breaking change?

Yeah, I don't think we can safely do that. It could be that there are applications out there that have these type codes hard coded, and changing the actual type code to something else in the driver could then lead to failures, as the driver checks that the type code that an application uses is valid.

I do however think that we can fix this in a different way:

1. The standard type code still remains the current 100,000 + internal code.
2. In addition, we'll define a type code that fits in a short. This type code will be the unchecked value of (short) VENDOR_TYPE_CODE. For JSON, this will be -31061.
3. We'll allow both the short vendor code and the normal vendor code to be used.
4. We'll remove the checked cast specifically for the result set that is returned by DatabaseMetadata#getTypeInfo(). This means that anyone calling ResultSet#getShort("DATA_TYPE") will get the short vendor code, and can freely use that with the driver.