Pyright Default None

[collindutter]

We make frequent use of this pattern to lazily initialize values:

griptape/griptape/drivers/prompt/openai_chat_prompt_driver.py

Lines 107 to 115 in f5503bd

	_client: openai.OpenAI = field(default=None, kw_only=True, alias="client", metadata={"serializable": False})
	@lazy_property()
	def client(self) -> openai.OpenAI:
	return openai.OpenAI(
	base_url=self.base_url,
	api_key=self.api_key,
	organization=self.organization,
	)

griptape/griptape/tokenizers/base_tokenizer.py

Lines 18 to 27 in f5503bd

	max_input_tokens: int = field(kw_only=True, default=None)
	max_output_tokens: int = field(kw_only=True, default=None)
	def __attrs_post_init__(self) -> None:
	if hasattr(self, "model"):
	if self.max_input_tokens is None:
	self.max_input_tokens = self._default_max_input_tokens()
	if self.max_output_tokens is None:
	self.max_output_tokens = self._default_max_output_tokens()

We default the field with a value of None, and lazily initialize it later on if the user did not provide their own, non-None, value.

Pyright's latest release seems to have patched this out.

Now we get many of these:

  /Users/collindutter/Documents/griptape/griptape/docs/examples/src/talk_to_a_webpage_1.py:41:9 - error: Argument of type "None" cannot be assigned to parameter "prompt_driver" of type "BasePromptDriver" in function "__init__"
    "None" is not assignable to "BasePromptDriver" (reportArgumentType)

Presumably the change came from:

Fixed bug that leads to inconsistent behavior in certain cases involving the evaluation of a call whose target signature involves a parameter with a default argument value, notably where the type of the default value isn't assignable to the declared type of its parameter.

[collindutter]

This is a mess, some of the errors are from 3p libraries:

griptape/griptape/tools/calculator/tool.py

Line 29 in f5503bd

return TextArtifact(numexpr.evaluate(expression))

Diagnostics:
1. Argument of type "None" cannot be assigned to parameter "out" of type "ndarray[Unknown, Unknown]" in function "evaluate"
     "None" is not assignable to "ndarray[Unknown, Unknown]" [reportArgumentType]

https://github.com/pydata/numexpr/blob/be928864031ede0bf466fca625881b7e50a210bc/numexpr/necompiler.py?#L896

griptape/griptape/drivers/prompt/amazon_sagemaker_jumpstart_prompt_driver.py

Line 96 in f5503bd

output_tokens = len(self.tokenizer.tokenizer.encode(generated_text))

Diagnostics:
1. Argument of type "None" cannot be assigned to parameter "truncation" of type "bool | str | TruncationStrategy" in function "encode"
     Type "None" is not assignable to type "bool | str | TruncationStrategy"
       "None" is not assignable to "bool"
       "None" is not assignable to "str"
       "None" is not assignable to "TruncationStrategy" [reportArgumentType]

https://github.com/huggingface/transformers/blob/222505c7e4d08da9095d12ddb72fb653f4b6da33/src/transformers/tokenization_utils_base.py#L2622

Going to sit on this for a couple days, maybe pyright will revert. Appears to be a final decision.