Summary

After upgrading Npgsql from 9.0.4 to 10.0.0, DbDataReader.GetColumnSchema()/GetColumnSchemaAsync() started reporting PostgreSQL array columns as untyped System.Array in DbColumn.DataType, whereas previously it was a strongly typed CLR array (e.g. System.Int32[]). This looks like a regression (or a breaking change) because the element type is known (int4[]) and Npgsql can map it.

Expected behavior

For a query returning int4[] (e.g. SELECT ARRAY[1,2,3]::int[]), schema[0].DataType should be typeof(int[]) (or at least preserve the element type information).

Actual behavior

• Npgsql 9.0.4: schema[0].DataType == typeof(int[])
• Npgsql 10.0.0: schema[0].DataType == typeof(Array)

Minimal repro

using Npgsql;

await using var connection = new NpgsqlConnection("Host=localhost;Port=5432;Database=postgres;Username=postgres;Password=postgres");
await using var command = connection.CreateCommand();
command.CommandText = "SELECT ARRAY[1,2,3]::int[]";
await connection.OpenAsync();
try
{
    await using var reader = await command.ExecuteReaderAsync();
    var schema = await reader.GetColumnSchemaAsync();
    Console.WriteLine("DbColumn.DataType: " + schema[0].DataType);
    Console.WriteLine("DbColumn.DataTypeName: " + schema[0].DataTypeName);
}
finally
{
    await connection.CloseAsync();
}

Environment

• Npgsql: 9.0.4 ✅ / 10.0.0 ❌
• PostgreSQL server version: 14
• .NET: net10

Questions

Is this change intentional in 10.x (ADO.NET compliance / design decision), or should this be considered a regression?
If intentional: what is the recommended way to reliably get the element CLR type from schema for array columns (without parsing DataTypeName)?