Python: Chroma, MongoDB, Pinecone and Weaviate filters raise NotImplementedError for 'in' with a list literal (and Pinecone 'not' raises TypeError) #14557

What happens

In the Python vector store connectors for Chroma, MongoDB Atlas, Pinecone and Weaviate, a filter lambda that uses in or not in with a list literal always fails with NotImplementedError: Unsupported AST node: <class 'ast.List'>.

The ast.In() and ast.NotIn() branches exist in each _lambda_parser, but the right-hand side can never be parsed, because none of these parsers has a case ast.List(). Qdrant has it (qdrant.py, case ast.List(): return [self._lambda_parser(elt) for elt in node.elts]), so the same filter works there.

Pinecone has a second problem on the same path. Its ast.Not() branch calls next(operand.values()), but dict.values() is a view and not an iterator, so any not (...) filter raises TypeError instead of the intended $nin rewrite or the intended NotImplementedError.

Where

python/semantic_kernel/connectors/ on current main (64005c7):

• chroma.py ChromaCollection._lambda_parser (case ast.In() / case ast.NotIn() at lines 382 and 384)
• mongodb.py MongoDBAtlasCollection._lambda_parser (lines 462 and 464)
• pinecone.py PineconeCollection._lambda_parser (lines 513 and 515)
• weaviate.py WeaviateCollection._lambda_parser (line 441)

Pinecone ast.Not() branch:

if (
    isinstance(operand, dict)
    and len(operand) == 1
    and isinstance(next(operand.values()), dict)  # type: ignore
    and "$in" in next(operand.values())  # type: ignore
):
    field = next(operand.keys())  # type: ignore
    values = next(operand.values())["$in"]  # type: ignore
    return {field: {"$nin": values}}

Redis (redis.py) also lacks case ast.List(), so it fails the same way. I left it out of the fix I am proposing because RedisVL Text and Num filters reject a list on the right side, so it needs a separate decision.

Steps to reproduce

from semantic_kernel.connectors.pinecone import PineconeCollection
from semantic_kernel.data.vector import VectorStoreCollectionDefinition, VectorStoreField

definition = VectorStoreCollectionDefinition(fields=[
    VectorStoreField("key", name="id"),
    VectorStoreField("data", name="tag", type="str", is_indexed=True),
    VectorStoreField("vector", name="vec", dimensions=2, type="list[float]"),
])
c = PineconeCollection.__new__(PineconeCollection)
object.__setattr__(c, "definition", definition)  # only _build_filter is exercised
c._build_filter("lambda x: x.tag in ['a', 'b']")
c._build_filter("lambda x: not (x.tag == 'a')")

The same call on ChromaCollection, MongoDBAtlasCollection and WeaviateCollection gives the same first error.

Actual

chroma    in ['a','b']      NotImplementedError: Unsupported AST node: <class 'ast.List'>
mongodb   in ['a','b']      NotImplementedError: Unsupported AST node: <class 'ast.List'>
pinecone  in ['a','b']      NotImplementedError: Unsupported AST node: <class 'ast.List'>
weaviate  in ['a','b']      NotImplementedError: Unsupported AST node: <class 'ast.List'>
pinecone  not (x.tag == 'a')   TypeError: 'dict_values' object is not an iterator
qdrant    in ['a','b']      FieldCondition(key='tag', match=MatchAny(any=['a', 'b']))

Expected

• x.tag in ['a', 'b'] becomes {"tag": {"$in": ["a", "b"]}} for Chroma, MongoDB and Pinecone, and contains_any for Weaviate. not in becomes $nin where the store supports it.
• Pinecone not (x.tag in [...]) becomes {"tag": {"$nin": [...]}}, and any other not expression raises the documented NotImplementedError("$not is only supported over $in ...").

I checked open and closed issues and PRs for these connectors and for filter parsing, and found nothing covering this. I am opening a PR right after filing this.