Question: Supporting Parquet datatypes

[ayushbindlish]

I have a use case where I need to generate data for parquet datatypes. I am currently using a custom version of JSF. Would you like to have this feature here?

JSON looks like the following:

"UInt32": {
      "type": "uint32"
    },
    "UInt64": {
      "type": "uint64"
    },
    "Float16": {
      "type": "float16"
    }

[number.py:jsf.src.schema_types.number:line 304 - generate()] - INFO: Generating random uint32 [number.py:jsf.src.schema_types.number:line 52 - generate()] - DEBUG: is_float: False [number.py:jsf.src.schema_types.number:line 72 - generate()] - INFO: Generated number: 35227457 [number.py:jsf.src.schema_types.number:line 333 - generate()] - INFO: Generating random uint64 [number.py:jsf.src.schema_types.number:line 52 - generate()] - DEBUG: is_float: False [number.py:jsf.src.schema_types.number:line 72 - generate()] - INFO: Generated number: 4669327448559716910 [number.py:jsf.src.schema_types.number:line 362 - generate()] - INFO: Generating random float16 [number.py:jsf.src.schema_types.number:line 57 - generate()] - DEBUG: is_float: True [number.py:jsf.src.schema_types.number:line 72 - generate()] - INFO: Generated number: 1.920763087895552e+17

[ghandic]

I don't believe JSON schema supports those numeric types, unless you can point me to the definition in the schema?

[ayushbindlish]

As you correctly mentioned, json schema does not support these datatypes but this is just for ease of data generation. For a given datatype, there are implicit ranges which can be manipulated using "minimum" and "maximum" but should always fall between the range for that datatype.

[ghandic]

I wouldn't add those types into jsf directly but if you proposed a PR for allowing people to extend the JSON Schema types with custom types it would work.

Then you would just manage your custom type generator classes outside of jsf

[ayushbindlish]

@ghandic Any ideas on how this can be done?

[ghandic]

Yes, unfortunately I don't have time to implement but PR's are welcome.

Would be along the lines of making subclasses of the given base class and defining a mapping for a type to the class that should be ran