[BUG] Passing None type to pyflyte run command is not supported

[0yukali0]

Flyte & Flytekit version

flyteidl 1.16.1
flytekit 1.16.9

Describe the bug

I passed None to pyflyte run to trigger task which has a optional parameter.
In local mode, TypeTransformerFailedError showed up.
In remote mode, there was an error with converter.

import flytekit as fl
from typing import Optional

@fl.task()
def hello(name: Optional[int] = None):
    print(name)

Expected behavior

pyflyte run command support --var1 None in local mode and remote mode.

Additional context to reproduce

import flytekit as fl
from typing import Optional

@fl.task()
def hello(name: Optional[int] = None):
    print(name)

1. local mode
   pyflyte run main.py hello --name None
2. remote mode
   pyflyte run --remote main.py hello --name None

Screenshots

1. Local mode

2. Remote mode

Are you sure this issue hasn't been raised already?

• Yes

Have you read the Code of Conduct?

• Yes

[0yukali0]

#take

[runllm]

Passing None to an optional parameter using pyflyte run (e.g., --name None) is not fully supported in all Flytekit versions. This is a known issue: Flytekit's type engine has historically struggled to serialize/deserialize None for Optional/Union types, leading to TypeTransformerFailedError or converter errors in both local and remote modes. This limitation is documented in several GitHub issues, and while there have been improvements (notably with the introduction of Union type support), edge cases remain problematic, especially for command-line None values ([BUG] Flyte seems to fail to deserialize serialized None to NoneType, [BUG] Non-primitive defaults to workflows, [BUG] OneHotEncoder type not supported as task input when None is supplied as input value, [BUG] Flytekit python unable to parse int | None style type hints on python3.11, [BUG] Local testing using optional dicts with optional values does not work).

As of Flytekit 1.16.x, there is no documented, fully reliable way to pass None via the CLI for optional parameters in both local and remote execution. Workarounds include using default values in code, or explicitly handling string "None" in your task logic, but these are not ideal. The Flyte team is aware and has discussed possible fixes and RFCs for better Union/Optional handling, but the CLI None-passing issue is not fully resolved (rfc/core language/sum-types.md, CHANGELOG-v1.1.0.md). Would you like more detail on workarounds or the current state of fixes?

Sources:

• [BUG] Flyte seems to fail to deserialize serialized None to NoneType when this type is eligible. #5573
• [flytekit] Non-primitive defaults to workflows #910
• [BUG] OneHotEncoder type not supported as task input when None is supplied as input value. #3319
• https://github.com/flyteorg/flyte/blob/master/rfc/core%20language/sum-types.md
• CHANGELOG-v1.1.0.md

---

This is an AI-generated response and your feedback is appreciated! Please leave a 👍 if this is helpful and 👎 if it is not.

Mention @RunLLM to ask follow-up questions. See runllm.com for more!

[machichima]

Hi @0yukali0 ,
Could you please provide a use case for this? When we will want to pass value None to an argument using pyflyte run. I think we can just not provide the argument (e.g.pyflyte run main.py hello), and the parameter name will then be None.
Thank you!

[0yukali0]

Hi, @machichima
I agree with your opinion.
This proposal just allow user to locally test a specific function where default value is not None and supported type includes None.

pyflyte run main.py encoding --data None
pyflyte run main.py encoding --data 1
pyflyte run main.py encoding --data False

import flytekit as fl
import typing

@fl.task()
def encoding(data: typing.Union[int, None, bool]  = 0) -> int:
    if isinstance(data, bool):
        return 2 if data else 3

    if isinstance(data, int):
        return data + 1

    if data is None:
        return 0

@fl.workflow
def map_workflow(data: list[typing.Union[int, None]] = [10, 12, None, 10, None, 12, 100, 11, 12, 10]) -> list[int]:
    return fl.map_task(encoding)(data=data)