fix(insights): coerce nullable string args of array-returning functions

splitByChar (and the other string-splitting functions) return Array(...).
Given a Nullable string argument — e.g. a breakdown on splitByChar('@',
properties.$distinct_id) — ClickHouse would produce Nullable(Array(...)),
which it rejects, failing the whole insight query.

The printer now wraps a nullable string argument of these functions in
ifNull(arg, '') so the result type stays a plain Array.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: sampennington

posthog/hogql/printer/clickhouse.py

@@ -71,6 +71,22 @@ def team_id_guard_for_table(table_type: ast.TableOrSelectType, context: HogQLCon
     )
 
 
+# String functions that return an Array — a Nullable string arg would produce an
+# illegal Nullable(Array(...)) type, so their nullable string args are coerced.
+_ARRAY_RETURNING_STRING_FUNCTIONS = frozenset(
+    {
+        "splitByChar",
+        "splitByString",
+        "splitByRegexp",
+        "splitByWhitespace",
+        "splitByNonAlpha",
+        "alphaTokens",
+        "extractAllGroups",
+        "ngrams",
+        "tokens",
+    }
+)
+
 # In non-nullable materialized columns, these values are treated as NULL
 MAT_COL_NULL_SENTINELS = ["", "null"]
 
@@ -155,6 +171,18 @@ def _render_function_call(self, node: ast.Call, func_meta) -> str:
                         args.append(f"ifNull({self.visit(arg)}, '')")
                 else:
                     args.append(f"ifNull(toString({self.visit(arg)}), '')")
+        elif node.name in _ARRAY_RETURNING_STRING_FUNCTIONS:
+            # These return Array(...). A Nullable string argument would make the
+            # result Nullable(Array(...)), which ClickHouse rejects, so coerce
+            # nullable string args to a non-nullable empty string.
+            args = []
+            for arg in node_args:
+                visited = self.visit(arg)
+                arg_type = arg.type.resolve_constant_type(self.context) if arg.type is not None else None
+                if isinstance(arg_type, ast.StringType) and arg_type.nullable:
+                    args.append(f"ifNull({visited}, '')")
+                else:
+                    args.append(visited)
         else:
             args = [self.visit(arg) for arg in node_args]
 

posthog/hogql/printer/test/test_printer.py

@@ -4813,6 +4813,12 @@ def test_can_call_parametric_function(self):
         )
         assert query_response.results == [(6,)]
 
+    def test_split_function_on_nullable_property(self):
+        # splitByChar on a nullable property would produce an illegal Nullable(Array)
+        query = parse_select("SELECT splitByChar('@', properties.email) FROM events")
+        query_response = execute_hogql_query(team=self.team, query=query)
+        assert query_response.results is not None
+
 
 class TestPostgresPrinter(BaseTest):
     maxDiff = None