fix: use flatten=False in Document.__eq__ to avoid metadata key collisions

[rekha0suthar]

Summary

Document.__eq__ called to_dict() with the default flatten=True, which merges the document's meta dictionary into the top-level representation. When a document's metadata contains a key that shares a name with a real Document field (such as "id" or "score"), the meta value silently overwrites the field value before the comparison is made. This causes two Documents with different metadata to compare as equal.

Example that was broken before this fix:

doc1 = Document(content="text", meta={"id": "meta-id-1"})
doc2 = Document(content="text", meta={"id": "meta-id-2"})
assert doc1 != doc2  # was incorrectly passing as equal

Fix

Changed line 97 of haystack/dataclasses/document.py from:

return self.to_dict() == other.to_dict()

to:

return self.to_dict(flatten=False) == other.to_dict(flatten=False)

flatten=False keeps the meta dictionary nested under its own key, so collisions between meta keys and top-level field names cannot affect the comparison result.

Tests

Added test_equality_with_colliding_meta_keys to test/dataclasses/test_document.py to prevent regressions on this exact case.

Fixes #11969

[vercel]

@rekha-suthar 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.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[github-actions]

Hi @rekha0suthar, thanks for your interest in contributing to Haystack! 🙏

⚠️ Issue #11969 is already being addressed by open pull request(s) #11982. Before opening a PR for an issue, please check whether a PR is already linked to it, and consider contributing to the existing PR instead. We may close duplicate PRs to keep the review queue manageable.

This is an automated message to help us keep the review queue healthy.

[rekha0suthar]

Closing as a duplicate of #11982 which is already addressing this issue. Thanks for the heads up!