Type annotations with on_predicate

[akaptur]

I'm writing some code that uses on_predicate to rate limit in an API client, roughly like this:

import backoff
from typing import TypeVar

_T = TypeVar("_T")

  @backoff.on_predicate(
      backoff.constant,
      lambda x: x is None,
  )
  def make_request(fn: Callable[..., _T]) -> Optional[_T]:  # This is the type signature I want
      if not is_allowed():  # rate limiting goes here
          return None

      return fn()

This function will always return the return value of fn(), since it continues trying forever - but because I'm using None as the sentinel object for the predicate, mypy wants the return type to be Optional[_T] instead of _T. I end up leaking implementation details to my callers, since None isn't ever returned to them, but the type system can't understand that.

Are there patterns that people use to avoid this problem? I could write a wrapper function that casts away the optional, or I could raise an exception and use the on_exception handler instead, but I'm curious what others have done.

[akaptur]

I ended up using backoff.on_exception instead of backoff.on_predicate and changing my implementation to raise instead of returning None. Since exceptions aren't captured in Python's type system, this lets me avoid the type signature problem.