Alexis Cubilla created CALCITE-7524:
---------------------------------------
Summary: JdbcSchema throws exception for DECIMAL columns with
precision 0 in JDBC metadata
Key: CALCITE-7524
URL: https://issues.apache.org/jira/browse/CALCITE-7524
Project: Calcite
Issue Type: Bug
Components: jdbc-adapter, jdbc-driver
Affects Versions: 1.41.0
Environment: Apache Calcite 1.41.0 or later (SqlTypeFactoryImpl
rejects DECIMAL precision 0).
PostgreSQL with columns defined as NUMERIC or DECIMAL without explicit
precision in DDL.
PostgreSQL JDBC driver (for example 42.x).
Reproduction path: JdbcSchema reads column metadata where COLUMN_SIZE is 0 for
DECIMAL/NUMERIC.
Reporter: Alexis Cubilla
Fix For: 1.42.0
Problem
-------
When using JdbcSchema with PostgreSQL (and other JDBC drivers), NUMERIC or
DECIMAL columns declared without explicit precision in DDL often report
COLUMN_SIZE = 0 in DatabaseMetaData.
JdbcSchema.sqlType() passes that value to createSqlType(DECIMAL, 0, scale).
SqlTypeFactoryImpl then rejects precision 0 with an error such as:
DECIMAL precision 0 must be between 1 and 19
The failure occurs while building JDBC column types (schema / row type), for
example when resolving a table that has such columns. It is not necessarily
caused by invalid end-user SQL.
Root cause
----------
Some JDBC drivers use column size 0 to mean that numeric precision is not
specified for unconstrained NUMERIC/DECIMAL. Calcite currently treats 0 as a
literal DECIMAL precision, which is invalid for the type factory.
Proposed fix
------------
In JdbcSchema.sqlType(), when the SQL type is DECIMAL and the reported
precision is 0, replace it with
typeFactory.getTypeSystem().getDefaultPrecision(DECIMAL) before calling
createSqlType.
Tests
-----
Add JdbcSchemaSqlTypeTest (or equivalent) covering Types.NUMERIC and
Types.DECIMAL with precision 0 and a positive scale, exercising the sqlType
mapping path.
Related issues
--------------
This is not the same as CALCITE-5212. CALCITE-5212 concerns
PRECISION_NOT_SPECIFIED (-1) and DECIMAL digest / DATATYPE_CACHE behavior. This
issue is only about literal precision 0 coming from JDBC metadata.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
