fix(datafusion): ensure that non-matching re_search calls return bool values when patterns do not match

Author: cpcloud

ibis/backends/datafusion/compiler/values.py

@@ -11,6 +11,7 @@
 import ibis.expr.datatypes as dt
 import ibis.expr.operations as ops
 from ibis.backends.base.sqlglot import (
+    FALSE,
     NULL,
     AggGen,
     F,
@@ -441,7 +442,16 @@ def string_find(op, *, arg, substr, start, end, **_):
 
 @translate_val.register(ops.RegexSearch)
 def regex_search(op, *, arg, pattern, **_):
-    return F.array_length(F.regexp_match(arg, pattern)) > 0
+    return if_(
+        sg.or_(arg.is_(NULL), pattern.is_(NULL)),
+        NULL,
+        F.coalesce(
+            # null is returned for non-matching patterns, so coalesce to false
+            # because that is the desired behavior for ops.RegexSearch
+            F.array_length(F.regexp_match(arg, pattern)) > 0,
+            FALSE,
+        ),
+    )
 
 
 @translate_val.register(ops.StringContains)

ibis/backends/tests/test_string.py

@@ -2,6 +2,7 @@
 
 import contextlib
 
+import numpy as np
 import pandas as pd
 import pytest
 import sqlalchemy as sa
@@ -1090,4 +1091,6 @@ def test_no_conditional_percent_escape(con, expr):
 )
 def test_non_match_regex_search_is_false(con):
     expr = ibis.literal("foo").re_search("bar")
-    assert con.execute(expr) is False
+    result = con.execute(expr)
+    assert isinstance(result, (bool, np.bool_))
+    assert not result