The Field mode normalization for null values is incorrect and sets it as an empty String.

Instead of setting the mode String as "" here, it should be set to Field.Mode.NULLABLE.name().

This causes the following problem: When the Field is deserialized from protobuf response from BigQuery, it might have mode set to null. Specifically this seems to happen when a table has been created with SQL CREATE TABLE as it doesn't support setting the field as NULLABLE explicitly.

Steps to reproduce

Create a BigQuery table with some nullable columns using CREATE TABLE.

Get the schema from Java with bigQuery.getTable().getSchema(), compare with a Schema you have created by setting the mode to Field.Mode.NULLABLE explicitly using Field.equals(field2).

Expected result: Both Schemas are equal.

Result: Schemas are inequal. This is because comparison is done for Protobuf representation where an empty String is a placeholder for null.

Code example

  @Test
  public void assertThatSchemaEqualsComparisonIsDeterministic() {
    final Field field1 = Field
      .of("field1", LegacySQLTypeName.INTEGER)
      .toBuilder()
      .setMode(Field.Mode.NULLABLE)
      .build();
    final Field field1b = Field
      .of("field1", LegacySQLTypeName.INTEGER)
      .toBuilder()
      .setMode(null)
      .build();
    final Schema schema1 = Schema.of(field1);
    final Schema schema2 = Schema.of(field1b);
    assertThat(schema1.equals(schema2)).isEqualTo(true);
  }