[C++][Python] Support Python-like slicing in list_slice kernel

Author: HyukjinKwon

cpp/src/arrow/compute/kernels/scalar_nested.cc

@@ -183,10 +183,9 @@ struct ListSlice {
     const auto* list_type = checked_cast<const BaseListType*>(list_array.type);
 
     // Pre-conditions
-    if (opts.start < 0 || (opts.stop.has_value() && opts.start >= opts.stop.value())) {
-      // TODO(ARROW-18281): support start == stop which should give empty lists
+    if (opts.start < 0 || (opts.stop.has_value() && opts.start > opts.stop.value())) {
       return Status::Invalid("`start`(", opts.start,
-                             ") should be greater than 0 and smaller than `stop`(",
+                             ") should be >= 0 and not greater than `stop`(",
                              ToString(opts.stop), ")");
     }
     if (opts.step < 1) {

cpp/src/arrow/compute/kernels/scalar_nested_test.cc

@@ -306,6 +306,32 @@ TEST(TestScalarNested, ListSliceOutputEqualsInputType) {
   }
 }
 
+TEST(TestScalarNested, ListSliceEmptyLists) {
+  // start == stop should return empty lists
+  auto input = ArrayFromJSON(list(int32()), "[[1, 2, 3], [4, 5], null]");
+  ListSliceOptions args(/*start=*/0, /*stop=*/0, /*step=*/1);
+  auto expected = ArrayFromJSON(list(int32()), "[[], [], null]");
+  CheckScalarUnary("list_slice", input, expected, &args);
+
+  // Different start position
+  args.start = 1;
+  args.stop = 1;
+  CheckScalarUnary("list_slice", input, expected, &args);
+
+  // Large list
+  auto input_large = ArrayFromJSON(large_list(int32()), "[[1, 2, 3], [4, 5]]");
+  args.start = 0;
+  args.stop = 0;
+  auto expected_large = ArrayFromJSON(large_list(int32()), "[[], []]");
+  CheckScalarUnary("list_slice", input_large, expected_large, &args);
+
+  // Fixed size list -> fixed size list[0]
+  auto input_fixed = ArrayFromJSON(fixed_size_list(int32(), 3), "[[1, 2, 3], [4, 5, 6]]");
+  args.return_fixed_size_list = true;
+  auto expected_fixed = ArrayFromJSON(fixed_size_list(int32(), 0), "[[], []]");
+  CheckScalarUnary("list_slice", input_fixed, expected_fixed, &args);
+}
+
 TEST(TestScalarNested, ListSliceBadParameters) {
   auto input = ArrayFromJSON(list(int32()), "[[1]]");
 
@@ -314,23 +340,14 @@ TEST(TestScalarNested, ListSliceBadParameters) {
                         /*return_fixed_size_list=*/true);
   EXPECT_RAISES_WITH_MESSAGE_THAT(
       Invalid,
-      ::testing::HasSubstr(
-          "`start`(-1) should be greater than 0 and smaller than `stop`(1)"),
+      ::testing::HasSubstr("`start`(-1) should be >= 0 and not greater than `stop`(1)"),
       CallFunction("list_slice", {input}, &args));
   // start greater than stop
   args.start = 1;
   args.stop = 0;
   EXPECT_RAISES_WITH_MESSAGE_THAT(
       Invalid,
-      ::testing::HasSubstr(
-          "`start`(1) should be greater than 0 and smaller than `stop`(0)"),
-      CallFunction("list_slice", {input}, &args));
-  // start same as stop
-  args.stop = args.start;
-  EXPECT_RAISES_WITH_MESSAGE_THAT(
-      Invalid,
-      ::testing::HasSubstr(
-          "`start`(1) should be greater than 0 and smaller than `stop`(1)"),
+      ::testing::HasSubstr("`start`(1) should be >= 0 and not greater than `stop`(0)"),
       CallFunction("list_slice", {input}, &args));
   // stop not set and FixedSizeList requested with variable sized input
   args.stop = std::nullopt;

python/pyarrow/tests/test_compute.py

@@ -3946,16 +3946,12 @@ def test_list_slice_field_names_retained(return_fixed_size, type):
 
 def test_list_slice_bad_parameters():
     arr = pa.array([[1]], pa.list_(pa.int8(), 1))
-    msg = r"`start`(.*) should be greater than 0 and smaller than `stop`(.*)"
+    msg = r"`start`(.*) should be >= 0 and not greater than `stop`(.*)"
     with pytest.raises(pa.ArrowInvalid, match=msg):
         pc.list_slice(arr, -1, 1)  # negative start?
     with pytest.raises(pa.ArrowInvalid, match=msg):
         pc.list_slice(arr, 2, 1)  # start > stop?
 
-    # TODO(ARROW-18281): start==stop -> empty lists
-    with pytest.raises(pa.ArrowInvalid, match=msg):
-        pc.list_slice(arr, 0, 0)  # start == stop?
-
     # Step not >= 1
     msg = "`step` must be >= 1, got: "
     with pytest.raises(pa.ArrowInvalid, match=msg + "0"):
@@ -3964,6 +3960,33 @@ def test_list_slice_bad_parameters():
         pc.list_slice(arr, 0, 1, step=-1)
 
 
+def test_list_slice_empty_lists():
+    # Test start == stop should return empty lists
+    arr = pa.array([[1, 2, 3], [4, 5, None], [6, None, None], None])
+    result = pc.list_slice(arr, 0, 0)
+    expected = pa.array([[], [], [], None], type=pa.list_(pa.int64()))
+    assert result.equals(expected)
+
+    # Test with different start positions
+    result = pc.list_slice(arr, 1, 1)
+    assert result.equals(expected)
+
+    result = pc.list_slice(arr, 2, 2)
+    assert result.equals(expected)
+
+    # Test with large_list
+    arr_large = pa.array([[1, 2, 3], [4, 5, None]], pa.large_list(pa.int64()))
+    result = pc.list_slice(arr_large, 0, 0)
+    expected_large = pa.array([[], []], pa.large_list(pa.int64()))
+    assert result.equals(expected_large)
+
+    # Test with fixed_size_list -> output is fixed_size_list[0]
+    arr_fixed = pa.array([[1, 2, 3], [4, 5, 6]], pa.list_(pa.int64(), 3))
+    result = pc.list_slice(arr_fixed, 0, 0)
+    expected_fixed = pa.array([[], []], pa.list_(pa.int64(), 0))
+    assert result.equals(expected_fixed)
+
+
 def check_run_end_encode_decode(value_type, run_end_encode_opts=None):
     values = [1, 1, 1, 2, 2, 1, 1, 1, 1, 1, 3, 3, 3, 3, 3, 3, 3, 3, 3]
     arr = pa.array(values, type=value_type)