Avro - code-generation approach vs non-code generation approach

Question

I'm new to Avro. The official documentation indicates that there are two possible approaches to using avro;

• With code generation - here classes are auto-generated from avro schema files by the avro compiler. These classes are then used in the application code.

• Without code generation - the schemas are directly parsed in the application. Instead of using auto-generated classes (as in the first approach), the application creates GenericRecord objects instead.

Question

• What are the pros and cons of each approach (eg: in terms of performance, ease of writing the application code, etc.?

Thanks in advance.