Eason09053360 opened a new pull request, #73120:
URL: https://github.com/apache/airflow/pull/73120
## Why
Two `TypeError` messages in the Task SDK name a value that is not the one
being rejected, so they point a person debugging at the wrong thing.
`expand_kwargs` validates a list element-by-element, but reports the
container's type:
```python
process.expand_kwargs([{"file": "a.csv"}, 5])
# TypeError: expected XComArg or list[dict], not list
```
The list is exactly what `expand_kwargs` wants — the `5` inside it is the
problem, and the message never mentions it. `git log -S` traces this to #25925,
which added the element loop by copying the existing `raise` and updating the
message text but not the `type(kwargs)` in it. The `elif` branch below it,
where `kwargs` really is the rejected value, was left correct.
`get_merged_defaults` reports `task_params` when it rejects
`task_default_args`:
```python
# with a valid params= argument
TypeError: default_args must be a mapping, got <class 'ParamsDict'> # an
internal wrapper the caller never passed
# with no params= argument at all
TypeError: default_args must be a mapping, got <class 'NoneType'>
```
Neither message names `list`, which is what the caller actually passed as
`default_args`.
## What
- `task-sdk/src/airflow/sdk/bases/operator.py` — `get_merged_defaults`
reports `task_default_args`.
- `task-sdk/src/airflow/sdk/bases/decorator.py`,
`definitions/mappedoperator.py`, `definitions/decorators/task_group.py` — the
element loop in all three `expand_kwargs` implementations now reports the
container **and** the element: `not list containing int`.
Reporting both matters because `str` and `bytes` are `Sequence`s and take
the element loop too. Naming only the element would describe
`expand_kwargs("abc")` as `not list[str]` and `expand_kwargs(b"ab")` as `not
list[int]`, both worse than today's `not str` / `not bytes`; `bytearray` and
`memoryview` behave the same way, so there is no short list of types to
special-case. Keeping the container type means no input is described with less
detail than before, and a `tuple` of dicts with one bad element now reads `not
tuple containing int` instead of the misleading `not tuple`.
These four error paths had no test coverage, which is how the
`expand_kwargs` slip survived since 2022. Each site gets a parametrized test
covering the rejected element, the rejected container (guarding the `elif`
branch that was already correct), and `str`/`bytes`.
---
##### Was generative AI tooling used to co-author this PR?
- [X] Yes — Claude Code (Opus 5)
Generated-by: Claude Code (Opus 5) following [the
guidelines](https://github.com/apache/airflow/blob/main/contributing-docs/05_pull_requests.rst#gen-ai-assisted-contributions)
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
