[isc-tdyar]

Summary

• Fixed order_by(asc("distance")) bug where SQLAlchemy requires a column expression object, not a string literal
• Extracted the distance expression to a variable (distance_expr) and used it in both session.query() and order_by()
• Removed the unused asc import
• Fixed pre-existing test_irisvector test that was using an incorrect assertion pattern

Problem

When calling similarity_search_with_score_by_vector(), the query fails with an error because SQLAlchemy's order_by() clause was receiving a string literal "distance" via asc("distance"), but SQLAlchemy requires an actual column expression object.

Solution

Extract the distance expression that is already labeled as "distance" and reference it directly in the order_by() clause:

# Build the distance expression for ordering
distance_expr = (
    self.distance_strategy(embedding).label("distance")
    if self.native_vector
    else self.table.c.embedding.func(
        self.distance_strategy, embedding
    ).label("distance")
)

# Use the expression in both query and order_by
session.query(self.table, distance_expr)
    .filter(filter_by)
    .order_by(distance_expr)  # Fixed: was .order_by(asc("distance"))

Test Plan

• Added regression test test_irisvector_similarity_search_with_score_by_vector that directly calls the affected method
• Fixed pre-existing test_irisvector test to use consistent assertion pattern
• Ran full test suite with real IRIS containers: 13 passed across all IRIS versions

Files Changed

• langchain_iris/vectorstores.py: Fix implementation + removed unused asc import
• tests/test_vectorstores.py: Added regression test + fixed pre-existing test

[grongierisc]

May be this issue is also solved in pr #6

[daimor]

ohh, hey, looks like I'll need to merge both changes

[daimor]

tests working here