umarbutler / persist-cache Goto Github PK

Attempting to cache a function call that contains an argument that is a list containing a dictionary raises a TypeError because, when attempting to determine whether a Python object is directly msgpackable, we check if it is a list and if it is, we check if the first element of that list is within the LISTED_SIGNATURES set which is a set of signatures indicating that certain types of sequences have been converted to lists.

bytes, bytearray and memoryview are being treated as absolutely directly msgpackable when, in fact, the msgspec decoder is not given bytes, bytearray and memoryview type hints and so is incapable of deserialising them. The decoder is not given such type hints because we already type hint strs and doing so prevents the other types from being decoded to, as described in this issue on the msgspec GitHub project.

As discussed in this issue on the openai-python GitHub project and this issue on the cpython GitHub project, inspect.iscoroutinefunction, which persist-cache relies upon to detect asynchronous functions and methods, is incapable of recognising openai.AsyncCompletions.create as an asynchronous method.