fix(udf): make udfs impure and avoid merging selects with impure functions

Author: cpcloud

ibis/backends/sql/rewrites.py

@@ -3,6 +3,7 @@
 from __future__ import annotations
 
 import operator
+import sys
 from collections.abc import Mapping
 from functools import reduce
 from typing import TYPE_CHECKING, Any
@@ -230,6 +231,9 @@ def complexity(node):
     def accum(node, *args):
         if isinstance(node, ops.Field):
             return 1
+        elif isinstance(node, ops.Impure):
+            # consider (potentially) impure functions maximally complex
+            return sys.maxsize
         else:
             return 1 + sum(args)
 

ibis/expr/operations/generic.py

@@ -189,15 +189,15 @@ class Impure(Value):
 
 
 @public
-class TimestampNow(Constant):
+class TimestampNow(Impure):
     """Return the current timestamp."""
 
     dtype = dt.timestamp
     shape = ds.scalar
 
 
 @public
-class DateNow(Constant):
+class DateNow(Impure):
     """Return the current date."""
 
     dtype = dt.date

ibis/expr/operations/udf.py

@@ -51,7 +51,7 @@ class InputType(enum.Enum):
 
 
 @public
-class ScalarUDF(ops.Value):
+class ScalarUDF(ops.Impure):
     @attribute
     def shape(self):
         if not (args := getattr(self, "args")):  # noqa: B009
@@ -65,7 +65,7 @@ def shape(self):
 
 
 @public
-class AggUDF(ops.Reduction):
+class AggUDF(ops.Reduction, ops.Impure):
     where: Optional[ops.Value[dt.Boolean]] = None