Closed cartermak closed 1 year ago
The pandas
dependency will need to be rolled back to Version ^1.1.5 (released in 2020) which is the last version that supported Python 3.6.
The so-called drop-in replacement for dataclasses
turns out to not be what it's made out to be. I can't get it to play nice with non-standard field types and/or the dataclasses_json
package's config
encoder
/decoder
mechanism. So the problem persists of finding a nice way to handle (de)serialization.
dataclasses
with attrs
. Outstanding work to go is testing with 3.6 and catching any other issues there.
Python 3.6 compatibility would reduce a lot of installation/virtual environment complexity for hosts where Python 3.6 is default. I originally dropped this idea because Python
dataclasses
, a core piece of the way Aerie CLI handles data, were only introduced in Python 3.7. However, Pydantic has a drop-in replacement fordataclasses
and Version 1.9.2 supports Python 3.6 officially.This would also enable us to replace
dataclasses_json
, which is not as well-supported/popular as Pydantic.