[Ganesh1729-ui]

Summary

Metadata.from_raw(..., validate=True) can escape its grouped validation errors when an unrecognized field is combined with a missing or invalid metadata_version. A misspelled field raises AttributeError; keys such as from_raw and from_email can also raise AttributeError or disappear from the collected errors.

Check whether each raw key names a validator before the conditional metadata-version age check. Collect keys without a validator as InvalidMetadata, including classmethod-name collisions. This lets the existing collector report both the metadata-version error and the unrecognized field.

Reproduction

from packaging.metadata import Metadata

Metadata.from_raw({"name": "example", "version": "1.0", "dynamc": ["Summary"]})

Before: AttributeError for dynamc escapes validation.

After: the documented ExceptionGroup contains InvalidMetadata for both metadata-version and dynamc.

The nine parametrized regressions cover a misspelled key and both classmethod-name collisions, each with a missing, invalid or valid metadata version. Eight fail on unchanged main; the valid-version misspelled-key case is an existing-behavior control.

Validation

• Linux Python 3.10 and 3.13: complete configured test sessions, 62,502 passed on each, with 100% line and branch coverage.
• Windows Python 3.13: 62,501 passed / one existing Linux-only skip.
• Property tests: 427 passed.
• Prescribed nox -s lint: all hooks, source/wheel build and Twine checks pass.
• Fresh installed wheel: nine malformed-input cases and two valid/lazy controls pass; strict distribution checks pass.

Fork validation verifies the eight expected failures on unchanged main, then runs the complete patched Linux sessions and lint checks. Its helper workflow is on a separate fork-only branch; this PR changes only the validator and its tests.

[Wkhanex]

AI-assisted validation at 3778cdf9bc26 on Python 3.12.14: I exercised missing, invalid (0.0), and valid (2.6) metadata versions with five unrecognized names: unknown_field, from_raw, from_email, _raw, and __dict__. All 15 cases produced grouped InvalidMetadata errors containing exactly the offending field plus the metadata-version error when applicable. The caller's input mapping remained unchanged. All 353 metadata tests passed.

Including _raw and __dict__ is a useful extension to the classmethod collisions in the added tests: these names also exist on the class but are not metadata validators. The descriptor lookup now rejects those without escaping the aggregation path.

[Ganesh1729-ui]

Thanks @Wkhanex for validating the descriptor-collision cases, including _raw and dict. That matches the intended guard: keys without a metadata validator are collected as InvalidMetadata instead of escaping or disappearing from the grouped validation errors.