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

Author: HyukjinKwon

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

@@ -162,7 +162,8 @@ Result<TypeHolder> ListSliceOutputType(const ListSliceOptions& opts,
           "`stop` being set.");
     }
     if (opts.step < 1) {
-      return Status::Invalid("`step` must be >= 1, got: ", opts.step);
+      return Status::Invalid("`step` must be greater than or equal to 1, got: ",
+                             opts.step);
     }
     const auto length = ListSliceLength(opts.start, opts.step, *stop);
     return fixed_size_list(value_type, static_cast<int32_t>(length));
@@ -183,14 +184,15 @@ 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
-      return Status::Invalid("`start`(", opts.start,
-                             ") should be greater than 0 and smaller than `stop`(",
-                             ToString(opts.stop), ")");
+    if (opts.start < 0 || (opts.stop.has_value() && opts.start > opts.stop.value())) {
+      return Status::Invalid(
+          "`start`(", opts.start,
+          ") should be greater than or equal to 0 and not greater than `stop`(",
+          ToString(opts.stop), ")");
     }
     if (opts.step < 1) {
-      return Status::Invalid("`step` must be >= 1, got: ", opts.step);
+      return Status::Invalid("`step` must be greater than or equal to 1, got: ",
+                             opts.step);
     }
 
     auto* pool = ctx->memory_pool();

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

@@ -176,6 +176,12 @@ TEST(TestScalarNested, ListSliceVariableOutput) {
   auto input = ArrayFromJSON(fixed_size_list(int32(), 1), "[[1]]");
   auto expected = ArrayFromJSON(list(int32()), "[[1]]");
   CheckScalarUnary("list_slice", input, expected, &args);
+
+  args.start = 0;
+  args.stop = 0;
+  auto input_empty = ArrayFromJSON(list(int32()), "[[1, 2, 3], [4, 5], null]");
+  auto expected_empty = ArrayFromJSON(list(int32()), "[[], [], null]");
+  CheckScalarUnary("list_slice", input_empty, expected_empty, &args);
 }
 
 TEST(TestScalarNested, ListSliceFixedOutput) {
@@ -315,22 +321,17 @@ TEST(TestScalarNested, ListSliceBadParameters) {
   EXPECT_RAISES_WITH_MESSAGE_THAT(
       Invalid,
       ::testing::HasSubstr(
-          "`start`(-1) should be greater than 0 and smaller than `stop`(1)"),
+          "`start`(-1) should be greater than or equal to 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)"),
+          "`start`(1) should be greater than or equal to 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;
@@ -343,9 +344,9 @@ TEST(TestScalarNested, ListSliceBadParameters) {
   args.start = 0;
   args.stop = 2;
   args.step = 0;
-  EXPECT_RAISES_WITH_MESSAGE_THAT(Invalid,
-                                  ::testing::HasSubstr("`step` must be >= 1, got: 0"),
-                                  CallFunction("list_slice", {input}, &args));
+  EXPECT_RAISES_WITH_MESSAGE_THAT(
+      Invalid, ::testing::HasSubstr("`step` must be greater than or equal to 1, got: 0"),
+      CallFunction("list_slice", {input}, &args));
 }
 
 TEST(TestScalarNested, StructField) {

python/pyarrow/tests/test_compute.py

@@ -3898,7 +3898,8 @@ def test_list_slice_output_fixed(start, stop, step, expected, value_type,
     (0, 1,),
     (0, 2,),
     (1, 2,),
-    (2, 4,)
+    (2, 4,),
+    (0, 0,)
 ))
 @pytest.mark.parametrize("step", (1, 2))
 @pytest.mark.parametrize("value_type", (pa.string, pa.int16, pa.float64))
@@ -3946,18 +3947,14 @@ 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 greater than or equal to 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: "
+    msg = "`step` must be greater than or equal to 1, got: "
     with pytest.raises(pa.ArrowInvalid, match=msg + "0"):
         pc.list_slice(arr, 0, 1, step=0)
     with pytest.raises(pa.ArrowInvalid, match=msg + "-1"):