DRY DataFrameModel fields

[lundybernard]

Given many data fields from various sources (raw input, computed, normalized, etc),
I need to create many DataFrameModels, composed of different sets of these fields.
I want to re-use field/column/DataFrameModel definitions to DRYup the code as much as possible.

I have had some success re-using DataFrameModels:

class F1(DataFrameModel):
    f1: Int = Field()


class F2(DataFrameModel):
    f2: Int = Field()
    
    
class Composite(F1, F2, DataFrameModel): ...

• Note: Actual code is more verbose, makes use of Field validation, and Config sub-classes

However, this approach breaks down if I try to build intermediate collections of fields

class Raw1(DataFrameModel):
    raw1: Int = Field()


class Raw2(DataFrameModel):
    raw2: Int = Field()
    
class RawData(Raw1, Raw2, DataFrameModel): ...

class Computed1(DataFrameModel):
    computed1: Float = Field()
    
class Computed2(DataFrameModel):
    computed2: Float = Field
    
class ComputedData(Computed1, Computed2, DataFrameModel): ...

# OK up to this point

class CollectedData(RawData, ComputedData, DataFrameModel): ...

This breaks down due to the classic diamond-shaped MRO problem

E   TypeError: Cannot create a consistent method resolution
E   order (MRO) for bases ...

I have attempted to refactor the code using Mixin classes, the standard solution to this problem, but that approach breaks the .to_schema() method.

class Raw1Mixin:
    raw1: Int = Field()
    
class Raw1(Raw1Mixin, DataFrameModel): ...

class Raw2Mixin:
    raw2: Int = Field()

class Raw2(Raw2Mixin, DataFrameModel): ...

    
class RawData(Raw1Mixin, Raw2Mixin, DataFrameModel): ...

This breaks at this point, even before we get to nested collections of fields.

RawData.to_schema()
...pandera/api/pandas/model.py:236: in to_schema
    cls.__fields__ = cls._collect_fields()
...pandera/api/pandas/model.py:449: in _collect_fields
    field = attrs[field_name]  # __init_subclass__ guarantees existence
E   KeyError: 'raw2'

Before delving any deeper into the inner-workings of _collect_fields I wanted to stop and ask for advice.

What is the best, most pythonic, most Pandera'ic, way to create DRY, structured, DataFrameModels like this?
Primary concerns:

• readability and maintainability
• keep Pandera's type-checking and type-hinting working.
• keep hypothesis .strategy() working