bug: Document.__eq__ ignores colliding metadata keys during equality comparison

[aryanp160]

Describe the bug

Document.__eq__ can incorrectly report two Document instances as equal when their metadata contains keys that collide with top-level Document fields (for example id, content, or score).

The current implementation compares the output of Document.to_dict(). Since to_dict() defaults to flatten=True, metadata is flattened into the top-level dictionary and any colliding metadata keys are overwritten by the document's own fields before the equality comparison takes place.

As a result, two documents with different metadata can compare as equal even though their logical state differs.

---

Error message

No exception is raised.

The issue is silent because the equality comparison simply returns the wrong result.

from haystack import Document

doc1 = Document(
    id="same_id",
    content="hello",
    meta={"id": "different1"},
)

doc2 = Document(
    id="same_id",
    content="hello",
    meta={"id": "different2"},
)

print(doc1 == doc2)

Output:

True

---

Expected behavior

Documents with different metadata should not compare as equal.

Expected output:

False

---

Additional context

While investigating this, I verified the following:

• Document.to_dict(flatten=False) preserves the metadata correctly.
• The collision only occurs in the flattened representation.
• Changing Document.__eq__ locally to compare the unflattened representation fixes the issue.
• I ran the full test/dataclasses/test_document.py suite after making this change, and all existing tests still passed.

I also have a regression test ready if this is considered a bug.

---

To Reproduce

from haystack import Document

doc1 = Document(
    id="same_id",
    content="hello",
    meta={"id": "different1"},
)

doc2 = Document(
    id="same_id",
    content="hello",
    meta={"id": "different2"},
)

assert doc1 != doc2

Current result:

doc1 == doc2
# True

Expected result:

doc1 == doc2
# False

---

FAQ Check

• Have you had a look at the FAQ page?

---

System

• OS: Windows 11
• Haystack version: Current main branch (verified locally)

[simrangarg482-bot]

Hi! I looked into this issue and was able to reproduce it locally.

It appears that Document.__eq__ compares the flattened output of to_dict(), which causes metadata keys like id to be overwritten by the document's top-level fields during comparison. Using the unflattened representation (flatten=False) seems to resolve the issue, and I also have a regression test prepared.

If no one is already working on this, I'd be happy to work on a fix and submit a PR.