Merge pull request #2859 from rgsl888prabhu/2495_tail_method_fail_on_…

…string

[REVIEW] Fix tail method issue for string columns

CHANGELOG.md

@@ -116,6 +116,7 @@
 - PR #2809 Add hash_df and group_split dispatch functions for dask
 - PR #2843 Parquet reader: fix skip_rows when not aligned with page or row_group boundaries
 - PR #2851 Deleted existing dask-cudf/record.txt
+- PR #2859 Fix tail method issue for string columns
 - PR #2852 Fixed `cumsum()` and `cumprod()` on boolean series.
 - PR #2750 Fixed casting values to cudf::bool8 so non-zero values always cast to true
 - PR #2873 Fixed dask_cudf read_partition bug by generating ParquetDatasetPiece

python/cudf/cudf/tests/test_dataframe.py

@@ -1847,6 +1847,13 @@ def check_frame_series_equality(left, right):
     check_frame_series_equality(gdf["a"].tail(-2), gdf["a"][2:])
 
 
+def test_tail_for_string():
+    gdf = DataFrame()
+    gdf["id"] = Series(["a", "b"], dtype=np.object)
+    gdf["v"] = Series([1, 2])
+    assert_eq(gdf.tail(3), gdf.to_pandas().tail(3))
+
+
 @pytest.mark.parametrize("drop", [True, False])
 def test_reset_index(pdf, gdf, drop):
     assert_eq(pdf.reset_index(drop=drop), gdf.reset_index(drop=drop))

python/nvstrings/nvstrings.py

@@ -419,12 +419,7 @@ def __getitem__(self, key):
         if isinstance(key, int):
             return self.gather([key])
         if isinstance(key, slice):
-            start = 0 if key.start is None else key.start
-            end = self.size() if key.stop is None else key.stop
-            step = 1 if key.step is None or key.step == 0 else key.step
-            # negative slicing check
-            end = self.size() + end if end < 0 else end
-            start = self.size() + start if start < 0 else start
+            start, end, step = key.indices(self.size())
             rtn = pyniNVStrings.n_sublist(self.m_cptr, start, end, step)
             if rtn is not None:
                 rtn = nvstrings(rtn)

python/nvstrings/tests/test_array.py

@@ -1,5 +1,7 @@
 # Copyright (c) 2018-2019, NVIDIA CORPORATION.
 
+from utils import assert_eq
+
 import nvstrings
 
 
@@ -44,3 +46,10 @@ def test_scalar_scatter():
     got = s1.scalar_scatter("+", [1, 3], 2)
     expected = ["a", "+", "c", "+"]
     assert got.to_host() == expected
+
+
+def test_slice_negative_start():
+    strs = nvstrings.to_device(["a", "b"])
+    got = strs[slice(-3, None, None)]
+    expected = ["a", "b"]
+    assert_eq(got, expected)