Replaceing the metadata parser from packaging.metadata

[LalatenduMohanty]

Fixes #561. Replacing the metadata parser with the metadata parser from packaging.metadata.
As we need to use the packaging library to parse metadata instead of using the email library ourselves.

[LalatenduMohanty]

@tiran @dhellmann PTAL

[dhellmann]

This looks good to me. How many places in the code do we do something similar to parse metadata? How useful would it be to have a function that takes a path and returns the metadata?

[tiran]

packaging.metadata.parse_email does not round trip and looses any field that it does not understand. Is this okay? Do we need a round trip-safe function?

[LalatenduMohanty]

How many places in the code do we do something similar to parse metadata? How useful would it be to have a function that takes a path and returns the metadata?

My bad. I should have checked all code to see if same pattern exists else where. I can see https://github.com/python-wheel-build/fromager/blob/main/src/fromager/candidate.py#L82 . I do not think we need a common function yet. PTAL and let me know.

[LalatenduMohanty]

packaging.metadata.parse_email does not round trip and looses any field that it does not understand. Is this okay? Do we need a round trip-safe function?

Let me get back to you on this.

[LalatenduMohanty]

@tiran Since fromager only reads metadata for dependency resolution and doesn't need to write it back, round trip safety isn't necessary. The benefits of type safety and validation from packaging.metadata outweigh the loss of unknown fields that aren't being used anyway.

[rd4398]

Looks good to me! I will wait for @tiran to approve since he had clarification questions

[tiran]

This does not work. A metadata object has three mandatory fields:

>>> raw_metadata, _ = parse_email(b"")
>>> Metadata.from_raw(raw_metadata)
  + Exception Group Traceback (most recent call last):
  |   File "<python-input-4>", line 1, in <module>
  |     Metadata.from_raw(raw_metadata)
  |     ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^
  |   File "/usr/lib/python3.13/site-packages/packaging/metadata.py", line 752, in from_raw
  |     raise ExceptionGroup("invalid metadata", exceptions)
  | ExceptionGroup: invalid metadata (3 sub-exceptions)
  +-+---------------- 1 ----------------
    | Traceback (most recent call last):
    |   File "/usr/lib/python3.13/site-packages/packaging/metadata.py", line 712, in from_raw
    |     metadata_version = ins.metadata_version
    |                        ^^^^^^^^^^^^^^^^^^^^
    |   File "/usr/lib/python3.13/site-packages/packaging/metadata.py", line 514, in __get__
    |     value = converter(value)
    |   File "/usr/lib/python3.13/site-packages/packaging/metadata.py", line 536, in _process_metadata_version
    |     raise self._invalid_metadata(f"{value!r} is not a valid metadata version")
    | packaging.metadata.InvalidMetadata: None is not a valid metadata version
    +---------------- 2 ----------------
    | Traceback (most recent call last):
    |   File "/usr/lib/python3.13/site-packages/packaging/metadata.py", line 747, in from_raw
    |     getattr(ins, key)
    |     ~~~~~~~^^^^^^^^^^
    |   File "/usr/lib/python3.13/site-packages/packaging/metadata.py", line 514, in __get__
    |     value = converter(value)
    |   File "/usr/lib/python3.13/site-packages/packaging/metadata.py", line 541, in _process_name
    |     raise self._invalid_metadata("{field} is a required field")
    | packaging.metadata.InvalidMetadata: 'name' is a required field
    +---------------- 3 ----------------
    | Traceback (most recent call last):
    |   File "/usr/lib/python3.13/site-packages/packaging/metadata.py", line 747, in from_raw
    |     getattr(ins, key)
    |     ~~~~~~~^^^^^^^^^^
    |   File "/usr/lib/python3.13/site-packages/packaging/metadata.py", line 514, in __get__
    |     value = converter(value)
    |   File "/usr/lib/python3.13/site-packages/packaging/metadata.py", line 554, in _process_version
    |     raise self._invalid_metadata("{field} is a required field")
    | packaging.metadata.InvalidMetadata: 'version' is a required field
    +------------------------------------

[tiran]

validate=False is not going to work either. It creates a broken Metadata object:

>>> raw_metadata, _ = parse_email(b"")
>>> m = Metadata.from_raw(raw_metadata, validate=False)
>>> m.name
Traceback (most recent call last):
  File "<python-input-4>", line 1, in <module>
    m.name
  File "/usr/lib/python3.13/site-packages/packaging/metadata.py", line 514, in __get__
    value = converter(value)
  File "/usr/lib/python3.13/site-packages/packaging/metadata.py", line 541, in _process_name
    raise self._invalid_metadata("{field} is a required field")
packaging.metadata.InvalidMetadata: 'name' is a required field

[tiran]

Why are you using parse_email() + Metadata.from_raw() instead of Metadata.parse_email()? The Metadata.parse_email() combines parse_email(), Metadata.from_raw(), and additional validation.

This code should probably use fromager.dependencies.parse_metadata(metadata_filename).

[tiran]

How about this?

Suggested change

	metadata_content = z.read(n)
	raw_metadata, _ = parse_email(metadata_content)
	metadata = Metadata.from_raw(raw_metadata)
	return metadata
	return Metadata.parse_email(z.read(n))

[tiran]

Why did you change this? Metadata.from_email() is better, because it performs additional validations with validate=True.