fix: recreate document id if certain attributes are changed

haystack/dataclasses/document.py

@@ -107,6 +107,16 @@ def __eq__(self, other):
             return False
         return self.to_dict() == other.to_dict()
 
+    def __setattr__(self, name, value):
+        # Only trigger custom logic if the field is already set (post-initialization)
+        if hasattr(self, name) or name == "meta":
+            super().__setattr__(name, value)
+            # Recreate document id if certain attributes are updated
+            if name in {"content", "dataframe", "blob", "meta", "embedding", "sparse_embedding"}:
+                object.__setattr__(self, "id", self._create_id())
+        else:
+            super().__setattr__(name, value)
+
     def __post_init__(self):
         """
         Generate the ID based on the init parameters.
@@ -122,7 +132,7 @@ def _create_id(self):
         dataframe = self.dataframe.to_json() if self.dataframe is not None else None
         blob = self.blob.data if self.blob is not None else None
         mime_type = self.blob.mime_type if self.blob is not None else None
-        meta = self.meta or {}
+        meta = self.meta if hasattr(self, "meta") and self.meta else {}
         embedding = self.embedding if self.embedding is not None else None
         sparse_embedding = self.sparse_embedding.to_dict() if self.sparse_embedding is not None else ""
         data = f"{text}{dataframe}{blob}{mime_type}{meta}{embedding}{sparse_embedding}"

releasenotes/notes/document-id-update-27d2356a791fc86e.yaml

@@ -0,0 +1,4 @@
+---
+fixes:
+  - |
+    Document ids are now updated upon changes of the document's attributes. Previously document ids were unchanged when adding meta data etc. to a document after its initialization

test/dataclasses/test_document.py

@@ -126,6 +126,34 @@ def test_basic_equality_id():
     assert doc1 != doc2
 
 
+@pytest.mark.parametrize(
+    "attribute, new_value",
+    [
+        ("content", "new content"),
+        ("dataframe", pd.DataFrame([10, 20, 30])),
+        ("blob", ByteStream(b"some bytes", mime_type="application/pdf")),
+        ("meta", {"date": "10-10-2023", "type": "article"}),
+        ("embedding", [0.1, 0.2, 0.3]),
+        ("sparse_embedding", SparseEmbedding(indices=[0, 2, 4], values=[0.1, 0.2, 0.3])),
+    ],
+)
+def test_id_is_updated_on_attribute_change(attribute, new_value):
+    doc1 = Document(content="test text")
+    id_at_creation = doc1.id
+
+    # Update attributes "content", "dataframe", "blob", "meta", "embedding", "sparse_embedding"
+    setattr(doc1, attribute, new_value)
+    assert doc1.id != id_at_creation
+
+
+def test_id_is_not_updated_on_score_change():
+    doc1 = Document(content="test text")
+    id_at_creation = doc1.id
+
+    doc1.score = 1.0
+    assert doc1.id == id_at_creation
+
+
 def test_to_dict():
     doc = Document()
     assert doc.to_dict() == {