Merge pull request #1338 from serhiizghama/fix/duckdb-bind-parameter-order

Keep DuckDB bind values in query order when a parameter repeats

src/python/txtai/database/duckdb.py

@@ -139,20 +139,16 @@ class DuckDB(Embedded):
             # Unpack query args
             query, parameters = args
 
-            # Iterate over parameters
+            # Match all named parameters in a single pass, so values are listed in the order they appear in the query
-            #   - Replace named parameters with ?'s
+            #   - Match on a word boundary, bind parameters can be followed by any non-word character, i.e. `in (:x)`
-            #   - Build list of value with position indexes
             params = []
-            for key, value in parameters.items():
+            if parameters:
-                # Match on a word boundary, bind parameters can be followed by any non-word character, i.e. `in (:x)`
+                pattern = r"\:(" + "|".join(re.escape(key) for key in parameters) + r")\b"
-                pattern = rf"\:{key}\b"
+                params = [parameters[match.group(1)] for match in re.finditer(pattern, query)]
-                for match in re.finditer(pattern, query):
-                    params.append((match.start(), value))
-
                 query = re.sub(pattern, "?", query)
 
             # Repack query and parameter list
-            args = (query, [value for _, value in sorted(params, key=lambda x: x[0])])
+            args = (query, params)
 
         return args
 

test/python/testdatabase/testrdbms.py

@@ -745,6 +745,13 @@ class Common:
             result = self.embeddings.search("select * from txtai where text in (:x, :y)", parameters={"x": self.data[1], "y": self.data[4]})
             self.assertEqual({row["text"] for row in result}, {self.data[1], self.data[4]})
 
+            # Test a repeated bind parameter followed by another bind parameter
+            result = self.embeddings.search(
+                "select * from txtai where text like :document or text like :document or text like :document or id = :id",
+                parameters={"document": "%iceberg%", "id": 4},
+            )
+            self.assertEqual({row["text"] for row in result}, {self.data[1], self.data[4]})
+
         def testSparse(self):
             """
             Test sparse vector search