[BUG] slice_strings producing incorrect results for some input strings

[jlowe]

Describe the bug
After #16574 cudf::strings::slice_strings is producing incorrect results for some inputs. See repro steps for details.

Steps/Code to reproduce bug
Apply the following patch and run gtests/STRINGS_TEST. The added test fails but will pass if #16574 is reverted.

diff --git a/cpp/tests/strings/slice_tests.cpp b/cpp/tests/strings/slice_tests.cpp
index 52e439bd93..bcadf5a7d4 100644
--- a/cpp/tests/strings/slice_tests.cpp
+++ b/cpp/tests/strings/slice_tests.cpp
@@ -50,6 +50,26 @@ TEST_F(StringsSliceTest, Substring)
   CUDF_TEST_EXPECT_COLUMNS_EQUAL(*results, expected);
 }
 
+TEST_F(StringsSliceTest, Substring_Failure)
+{
+  cudf::test::fixed_width_column_wrapper<uint8_t> char_data{
+    0x20,0x13,0x54,0x64,0xc3,0xb2,0x57,0x26,0x70,0x64,0xc2,0x93,0x67,0x00,0xc3,0xb4,0xc2,0xa4,0xc3,0xa3,0x5d,0x37,0xc2,0x80,0x2b,0x3a,0xc2,0xb9,0x73,0x77,0xc2,0xb7,0x16,0xc2,0xa9,0x3a,0xc3,0xa1,0x2c,0x60,0x70,0x2c,0x20,0xc2,0x80,0x56,0x58,0xc3,0xa2,0xc3,0xaf,0x0a,0xc2,0xa9,0x06,0x1e,0xc3,0xae,0xc3,0xb1,0x71,0xc2,0x8f,0x7f,0x42,0xc2,0x8b,0xc2,0x9b,0x24,0x6e,0x17,0xc3,0x98,0xc2,0x98,0xc3,0x9d,0xc3,0x9e,0xc2,0xbc,0xc2,0xb0,0x14,0x4d,0xc2,0xa1,0xc3,0x9c,0x2c,0x20,0xc2,0xb2,0xc2,0x9a,0x6a,0xc3,0x84,0xc2,0xaf,0x72,0xc2,0xb6,0xc3,0x86,0x24,0x1c,0x08,0x1f,0x07,0x4f,0x7b,0x15,0xc2,0xb5,0xc3,0xb9,0xc3,0x84,0x46,0xc2,0xa7,0x19,0x37,0xc3,0x96,0xc3,0xac,0xc2,0x88,0x0f,0xc3,0xac,0xc3,0x95,0x2e,0x2c,0x20,0xc3,0x8a,0xc2,0x90,0x2b,0xc2,0x9f,0xc2,0xac,0x6e,0xc2,0xa0,0xc3,0xaa,0xc3,0xa3,0xc3,0x94,0x6f,0x7e,0xc3,0xa8,0x7d,0x3a,0xc3,0xbd,0xc3,0x99,0x57,0x11,0x63,0x6b,0xc3,0x91,0x15,0xc3,0xb4,0x33,0x5e,0x62,0xc2,0x8e,0xc2,0x99,0xc3,0x9c,0x2c,0x20,0x1d,0x47,0xc2,0x8e,0x27,0xc2,0x98,0xc2,0x82,0xc2,0x93,0x37,0xc2,0xa2,0xc3,0x97,0xc3,0xa0,0xc3,0xad,0xc2,0x94,0x08,0x39,0xc3,0x94,0xc3,0xad,0xc2,0xbc,0x2e,0x51,0x2d,0x73,0x79,0xc2,0x89,0x1a,0xc3,0x9d,0xc2,0x8b,0xc2,0xad,0xc2,0x89,0x01,0x2c,0x20,0x59,0x0d,0xc3,0x99,0x56,0xc2,0x85,0xc3,0x84,0xc2,0x88,0x58,0x25,0xc3,0x91,0x38,0xc2,0x9f,0xc3,0x9b,0x00,0xc3,0xa2,0xc2,0xbd,0xc2,0x82,0xc2,0x80,0xc2,0xa7,0xc2,0xbb,0x72,0xc2,0xb4,0xc3,0x8d,0x12,0x48,0xc2,0x80,0xc3,0x8c,0xc3,0xbe,0xc3,0x95,0x1d,0x2c,0x20,0x1a,0x45,0xc2,0xbd,0xc2,0xb9,0xc2,0x8a,0x6b,0xc3,0x8a,0xc2,0x9a,0x79,0xc2,0x98,0x3a,0xc2,0xb5,0x50,0x73,0xc2,0xb3,0x71,0xc2,0x85,0xc2,0xbf,0x72,0x1b,0xc2,0x95,0x47,0xc3,0x84,0xc3,0xb2,0xc3,0xbd,0x36,0xc2,0xb1,0xc3,0xab,0xc3,0x9e,0x74,
+    0x20,0xc3,0x8e,0xc2,0x8b,0xc3,0xbe,0xc3,0xba,0xc3,0x93,0xc3,0x95,0x61,0x5c,0xc2,0xa4,0x4b,0x70,0x1f,0xc3,0x96,0x03,0x4c,0x72,0x0e,0xc3,0xbe,0x1b,0xc3,0x8c,0xc2,0x82,0xc2,0x8f,0xc3,0xa2,0xc2,0x85,0x16,0xc3,0x82,0xc3,0x97,0x1c,0x28,0xc2,0xac};
+  cudf::test::fixed_width_column_wrapper<int32_t> offsets{0, 334, 382};
+  cudf::column_view strs{
+    cudf::data_type{cudf::type_id::STRING},
+    2,
+    cudf::column_view(char_data).data<uint8_t>(),
+    nullptr,
+    0,
+    0,
+    {offsets}
+  };
+  int start = 0;
+  auto results = cudf::strings::slice_strings(strs, start);
+  CUDF_TEST_EXPECT_COLUMNS_EQUAL(*results, strs);
+}
+
 class Parameters : public StringsSliceTest, public testing::WithParamInterface<cudf::size_type> {};
 
 TEST_P(Parameters, Substring)

The failure looks like this. Note that the last character of the second input string was mangled in the output.

/cudf/cpp/tests/utilities/column_utilities.cu:562: Failure
Failed
first difference: lhs[1] =  ÎþúÓÕa\¤KpÖLrþâ
Â×(�, rhs[1] =  ÎþúÓÕa\¤KpÖLrþâ
Â×(¬
Google Test trace:
/cudf/cpp/tests/strings/slice_tests.cpp:70:  <--  line of failure

Apologies for the unfriendly random input data, but this is what was captured from a RAPIDS Spark integration test that was failing, and trying to narrow the input to just the failing string oddly does not fail.

Expected behavior
Test passes as it did before.

[davidwendt]

Closed by #16777