Major Description
In [9]: import avro.schema
In [10]: s = avro.schema.parse('{"type": "record", "name": "X", "fields": [{"name": "y", "type": {"type": "record", "name": "Y", "fields": [{"name": "Z", "type": "X"}]}}]}')
In [11]: str(s.fields[0].type)
Out[11]: '{"fields": [{"type": "X", "name": "Z"}], "type": "record", "name": "Y"}'
str(schema) is used in avro data files to record the schema. In the case above, when we serialize the schema for Y, we should actually also serialize the schema for X, since Y needs the schema for X.
I ran smack into this when using a schema from a protocol to write a data file, and finding that a lot of the types weren't defined when looking at the avro data file generated.