Please take a look at AVRO-3816. I'm proposing a backwards incompatible change to the python schema implementation. So far I have a PR in draft that I believe _is_ backwards-compatible, but is just a step on the way - it doesn't solve the problem entirely.
I'd like to get to the following advantages: 1. The standard and canonical properties are typed, able to benefit both core developers and api users. 2. The standard and canonical properties are slotted (https://docs.python.org/3/reference/datamodel.html#slots), providing a space savings when only standard attributes are in use. 3. Custom properties are part of the schema instance's __dict__ (dunder-dict) attribute, so standard python introspection tools like dir() and attribute access just work. To clarify, I don't think this will be backwards incompatible for people who just use the python implementation. The only incompatibility here is that I'm proposing to get rid of a handful of public methods like get_prop and set_prop, and change the underlying implementation. If any API users are directly accessing those methods, that will need to change. What do you think?
