Hi @Retired_mod 

Thanks for your help!
Your solution works for the initial issue,
and I've implemented it first in my code.

but it creates a other problem.
When we explicitly define the struct hint as 'bbb STRUCT<ccc: INT>',
it works until someone adds more fields to 'bbb'.

For example, with this data:
```python
data_file = [{"aaa": "2020-09-22T00:00:00Z", "bbb": {"ccc": 1234, "eee": "blabla"}, "ddd": "blabla"}]
```
Using these Schema Hints:
```python
SCHEMA_HINTS = [
'aaa TIMESTAMP',
'bbb STRUCT<ccc: INT>',
'ddd STRING'
]
```
We get an error because it can't handle additional fields in 'bbb' that are not specified in the hint:
org.apache.spark.sql.catalyst.util.UnknownFieldException:
```python
[UNKNOWN_FIELD_EXCEPTION.NEW_FIELDS_IN_RECORD_WITH_FILE_PATH] Encountered unknown fields during parsing: {"bbb":{"eee":"blabla"}}, which can be fixed by an automatic retry: false
```
The original Schema Hints we started with:
```python
SCHEMA_HINTS = [
'aaa TIMESTAMP',
'bbb.ccc INT'
]
```
do not have this problem and will add 'eee' if it exists in the data.

Currently, to work around the issue,
we've implemented a temporary solution.
We generate an initial data file that includes all nested fields specified in the Schema Hints,
such as 'bbb', and always write it to our source directory.
This file is then discarded after schema creation.

However, I'd love to hear if there's a better solution that addresses the problem more elegantly.
Thank you very much!