Handling optional fields and None values for HashModels

[kantholtz]

I was unable to find an explanation of how None values are handled. I've looked through the docs and hashmodel tests and did not find any explicit things regarding None values.

>>> from redis_om import HashModel, Field, Migrator                                                  
...                                                                                                  
... class A(HashModel):                       
...     x: str | None = Field(index=True)                                                                                                                                                                 
...                                                                                                  
... a = A(x=None)                                                                                                                                                                                         
>>> from redis_om import HashModel, Field, Migrator                                                  
...                                                                                                                                                                                                       
... class A(HashModel):
...     x: str | None = Field(index=True)     
...                                               
... a = A(x=None)
>>> a.save()
A(pk='01H9N93N9RFHHCQZS59P91572Y', x=None)
>>> A.get(a.pk)
A(pk='01H9N93N9RFHHCQZS59P91572Y', x='')

I think this is an unexpected behaviour (even if it makes sense when thinking from a "low-level" redis pov). It is also inconsistent with redis-py where saving a None value leads to str(None) being called which saves the literal string 'None' to redis.

See also #38, #254

I've opened this issue to discuss a fix (the other open issues have been open for a very long time) and I'm happy to help out code-wise. I have a few questions:

1. Is this intended behaviour? If so, a disclaimer should be added to the documentation and this issue can be closed.
2. It works for JsonModel instances. Is the use of the HashModel discouraged in general?
3. If it should be changed, how should it be approached? Obviously, assuming an empty string is always meant to be None is a bad idea. Maybe a "magic" (unlikely, random) string could be used instead?

[slorello89]

Hi @kantholtz - IMO this isn't the correct behavior, I wrote up a little bit more about the reason why this was the case in #616, and why now with pydantic 2.x support this is now fully broken (pending the resolution of that same PR :)) frankly, the default should be the default, and None's should be respected, and that's quite doable. The "Redis Way" to accomplish this is simply to exclude all None fields from the hash (which is what that PR will do).