Metadata object issues unwarranted deprecation warning

[Dev-iL]

Apache Airflow version

2.11.0

If "Other Airflow 2 version" selected, which one?

No response

What happened?

Following the documentation on Attaching extra information to an emitting dataset event causes an incorrect warning to be emitted. (The warning is incorrect in the sense that it says the user did something they didn't actually do - which is pass a string as the last input.)

Code from the docs' example:

    # Note that 3rd input, `alias`, isn't specified so it defaults to None.
    yield Metadata(example_s3_dataset, {"row_count": len(df)})

Then, in airflow/datasets/metadata.py, the branch that generates the warning is followed:

        if isinstance(alias, DatasetAlias):
            self.alias_name = alias.name
        else:  # This catches both `str` and `None`
            warnings.warn(...

What you think should happen instead?

The warning shouldn't say that a string (dataset uri) was passed instead of a Dataset object. Possibly, no warning should be emitted at all.

How to reproduce

Run the example from the docs verbatim.

Operating System

Linux

Versions of Apache Airflow Providers

No response

Deployment

Virtualenv installation

Deployment details

No response

Anything else?

Workaround

Provide a DatasetAlias(None) as the 3rd input to Metadata.

Are you willing to submit PR?

• Yes I am willing to submit a PR!

Code of Conduct

• I agree to follow this project's Code of Conduct

[BBQing]

I think the presence of warnings is correct. It warns about breaking changes coming in future releases.

@Dev-iL , you see the problem in how are the warnings phrased? Could you maybe provide some example, how they should be phrased? Probably include the None option in the alias branch?

[Dev-iL]

I think the presence of warnings is correct. It warns about breaking changes coming in future releases.

There's nothing wrong about a warning being issued in response to a deprecated behavior by the user, but this is not the case here.

you see the problem in how are the warnings phrased? Could you maybe provide some example, how they should be phrased?

No, the warning is fine for the use case it covers.

Probably include the None option in the alias branch?

That won't work, because a None object doesn't have a .name field. It needs to be a separate branch, e.g.

        if isinstance(alias, DatasetAlias):
            self.alias_name = alias.name
        elif alias is None:
            self.alias_name = None
        elif isinstance(alias, str):
            warnings.warn(...
        else:  # This handles any other data type
            raise TypeError

[BBQing]

I have meant to include the None case into the warning message to make it more clear, that even default value for alias parameter would cause error in the future.

[Dev-iL]

I have meant to include the None case into the warning message to make it more clear, that even default value for alias parameter would cause error in the future.

Oh is that the plan? That makes little sense becase it forces the users to use aliases when all they need is a plain dataset with some extras.

[BBQing]

I have meant to include the None case into the warning message to make it more clear, that even default value for alias parameter would cause error in the future.

Oh is that the plan? That makes little sense becase it forces the users to use aliases when all they need is a plain dataset with some extras.

Now I get it. You want to change it to something like this?

if isinstance(alias, DatasetAlias):
    self.alias_name = alias.name
else:
   if isinstance(alias, str):
        # issue warning
        ...
   self.alias_name = alias