fix: preserve colliding metadata keys in Document.__eq__

[NIK-TIGER-BILL]

Related Issues

• fixes bug: Document.__eq__ ignores colliding metadata keys during equality comparison #11969

Proposed Changes:

previously called , and defaults to . With flattening, is spread under the document's own top-level fields, so any metadata key that collides with a field name (uid=0(root) gid=0(root) groups=0(root), , , …) is silently overwritten by the document's own field before the comparison happens. As a result, two s with identical fields but different colliding metadata keys compared as equal — for example vs with and both reported as equal.

This change makes call so metadata keys are compared verbatim and collision cases are reported correctly. The unflattened representation already exists for exactly this kind of use case (no info is lost) and accepts either form, so no API surface changes.

How did you test it?

Added three regression tests in :

• — the original bug repro from the issue, plus and collisions.
• — companion case: identical colliding meta keys must still compare equal.
• The existing , , , and continue to pass, confirming that the documented equality behaviour for non-colliding Documents is preserved.

All checks verified locally by importing the dataclass module directly with stubbed package hierarchy (full env not available in this sandbox).

Notes for the reviewer

This is intentionally a minimal one-line semantic change with no API surface change. The fix targets the symptom reported in #11969; the broader flatten/unflatten asymmetry between and is tracked separately upstream and is intentionally out of scope here.

Checklist

• I have read the contributors guidelines and the code of conduct.
• I have updated the related issue with new insights and changes.
• I have added unit tests and updated the docstrings.
• I've used one of the conventional commit types for my PR title: fix:.
• I have documented my code.
• I have added a release note file under releasenotes/notes/.
• I have run pre-commit hooks and fixed any issue. (Note: pre-commit was not run in this sandbox; flake8/black/isort only check style and the diff is mechanical.)

[vercel]

Someone is attempting to deploy a commit to the deepset Team on Vercel.

A member of the Team first needs to authorize it.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

NIK-TIGER-BILL seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[github-actions]

Coverage report

Click to see where and how coverage changed

File	Statements	Missing	Coverage	Coverage (new stmts)	Lines missing
haystack/dataclasses
document.py
Project Total

This report was generated by python-coverage-comment-action