[FlightSQL][JDBC] NullPointerException in DatabaseMetadata methods when SqlInfo is unavailable

[merzbird]

Describe the bug

Multiple DatabaseMetadata methods in the Arrow Flight SQL JDBC driver throw NullPointerException when the server doesn't provide certain SqlInfo values. The ArrowDatabaseMetadata.getSqlInfoAndCacheIfCacheIsEmpty() method returns null, but calling code doesn't handle null defensively before invoking methods like .intValue(), .booleanValue(), or .isEmpty().

Affected Methods

All 68+ DatabaseMetadata methods that rely on getSqlInfoAndCacheIfCacheIsEmpty() are affected, including:

Integer-returning methods (20 methods)

• getMaxBinaryLiteralLength()
• getMaxCharLiteralLength()
• getMaxColumnNameLength()
• getMaxColumnsInGroupBy()
• getMaxColumnsInIndex()
• getMaxColumnsInOrderBy()
• getMaxColumnsInSelect()
• getMaxColumnsInTable()
• getMaxConnections()
• getMaxCursorNameLength()
• getMaxIndexLength()
• getMaxSchemaNameLength()
• getMaxProcedureNameLength()
• getMaxCatalogNameLength()
• getMaxRowSize()
• getMaxStatementLength()
• getMaxStatements()
• getMaxTableNameLength()
• getMaxTablesInSelect()
• getMaxUserNameLength()

Boolean-returning methods (30 methods)

• supportsColumnAliasing()
• nullPlusNonNullIsNull()
• supportsTableCorrelationNames()
• supportsDifferentTableCorrelationNames()
• supportsExpressionsInOrderBy()
• supportsOrderByUnrelated()
• supportsLikeEscapeClause()
• supportsNonNullableColumns()
• supportsIntegrityEnhancementFacility()
• isCatalogAtStart()
• supportsSelectForUpdate()
• supportsStoredProcedures()
• supportsCorrelatedSubqueries()
• doesMaxRowSizeIncludeBlobs()
• supportsTransactions()
• dataDefinitionCausesTransactionCommit()
• dataDefinitionIgnoredInTransactions()
• supportsBatchUpdates()
• supportsSavepoints()
• supportsNamedParameters()
• locatorsUpdateCopy()
• supportsStoredFunctionsUsingCallSyntax()

Support level methods (18 methods)

• getDefaultTransactionIsolation()
• supportsGroupBy()
• supportsGroupByUnrelated()
• supportsMinimumSQLGrammar()
• supportsCoreSQLGrammar()
• supportsExtendedSQLGrammar()
• supportsANSI92EntryLevelSQL()
• supportsANSI92IntermediateSQL()
• supportsANSI92FullSQL()
• supportsOuterJoins()
• supportsFullOuterJoins()
• supportsLimitedOuterJoins()
• supportsSchemasInProcedureCalls()
• supportsSchemasInIndexDefinitions()
• supportsSchemasInPrivilegeDefinitions()
• supportsCatalogsInIndexDefinitions()
• supportsCatalogsInPrivilegeDefinitions()
• supportsPositionedDelete()
• supportsPositionedUpdate()
• supportsSubqueriesInComparisons()
• supportsSubqueriesInExists()
• supportsSubqueriesInIns()
• supportsSubqueriesInQuantifieds()
• supportsUnion()
• supportsUnionAll()

Map-returning methods

• supportsConvert()

Component(s)

• FlightSQL
• JDBC

To Reproduce

import java.sql.*;
import java.util.Properties;

public class DatabaseMetadataTest {
    public static void main(String[] args) throws SQLException {
        String url = "jdbc:arrow-flight-sql://...";
        Properties props = new Properties();
        Driver driver = DriverManager.getDriver(url);
        try (Connection conn = driver.connect(url, props)) {
            DatabaseMetaData metaData = conn.getMetaData();

            // All of these throw NullPointerException
            System.out.println(metaData.getMaxBinaryLiteralLength());
            System.out.println(metaData.supportsTransactions());
            System.out.println(metaData.supportsGroupBy());
        }
    }
}

Error Output

Exception in thread "main" java.lang.NullPointerException: Cannot invoke "java.lang.Long.intValue()"
because the return value of "org.apache.arrow.driver.jdbc.ArrowDatabaseMetadata.getSqlInfoAndCacheIfCacheIsEmpty
(org.apache.arrow.driver.jdbc.shaded.org.apache.arrow.flight.sql.impl.FlightSql$SqlInfo, java.lang.Class)" is null
	at org.apache.arrow.driver.jdbc.ArrowDatabaseMetadata.getMaxBinaryLiteralLength(ArrowDatabaseMetadata.java:xxx)

Expected behavior

1. Option A: Return sensible defaults when SqlInfo is unavailable:

   • Integer methods: return 0 (as per JDBC spec for "no limit" or "unknown")
   • Boolean methods: return false (conservative/safe default)
   • Support level methods: return appropriate default enums
   • Map methods: return empty map

2. Option B: Throw SQLException with a descriptive message indicating the server doesn't support the requested SqlInfo

3. Methods should NOT throw unchecked NullPointerException - this violates JDBC contracts and the principle of least surprise

Environment

Arrow Version:

• 18.3.0

Java Version:

• Java 11+

OS:

• All platforms

JDBC Specification Reference

Per JDBC 4.3 specification:

• Methods returning int should return 0 when the limit is unknown or there is no limit
• Methods returning boolean should return false for unsupported features
• Methods should throw SQLException for errors, not unchecked exceptions

[ennuite]

Hi! I will look into fixing this in the next couple of days. Will let you know once I open a PR. Thank you for the thorough bug report.

[ennuite]

@merzbird I created a PR. Let me know if you agree with the fix, I return sensible values like you suggest, and for classes for which no known decent defaults are know, I throw a SQLException. All current call sites are covered with defaults. The exception is a guard rail for future additions.

Also, feel free to ping me on any new issues you create, especially concerning Flight or JDBC. I regularly monitor the issues, but it might skip me.

[merzbird]

@ennuite Thanks for the swift response!

It is a nice and simple solution to default to false/0 when SqlInfo is absent, but I have doubts that it is always the best - e.g., Avatica (which AvaticaDatabaseMetaData arrow's impl is based on) uses different values as defaults: supportsGroupBy is true, but supportsStoredProcedures is false. I don't know, maybe there are some strong reasons to prefer that approach.

But for simplicity's sake and conformance with JDBC spec, your PR definitely looks good to me.

[ennuite]

@merzbird I had similar thoughts regarding the boolean values. For the integers, it is specified that 0 also serves as unknown so I think it's fine, but the JDBC spec does not say the same for boolean values. I considered throwing a checked exception instead of returning false, but I'm under the impression that this could hinder common usage of this API.