[BUG][DATAGEN_SEED=1753338612] test_array_slice_with_negative_length failed in scala 2.13 integration

[abellina]

Given this part of the error, it looks to be we expect the CPU to throw an error and it is not (I think). It could be a "shape of data" problem as @firestarman mentions here: #12955.

[2025-07-24T11:27:04.911Z] ../../src/main/python/asserts.py:707: in assert_gpu_and_cpu_error
[2025-07-24T11:27:04.911Z]     assert_spark_exception(lambda: with_cpu_session(df_fun, conf), error_message)

[2025-07-24T11:27:04.910Z] __________ test_array_slice_with_negative_length[Array(Integer)--5-b] __________

[2025-07-24T11:27:04.910Z] [gw1] linux -- Python 3.10.18 /opt/conda/bin/python

[2025-07-24T11:27:04.910Z] 

[2025-07-24T11:27:04.910Z] data_gen = Array(Integer), valid_start = 'b', negative_length = -5

[2025-07-24T11:27:04.910Z] 

[2025-07-24T11:27:04.910Z]     @pytest.mark.parametrize('valid_start', [5, 'b'], ids=idfn)

[2025-07-24T11:27:04.910Z]     @pytest.mark.parametrize('negative_length', [-5, 'c'], ids=idfn)

[2025-07-24T11:27:04.910Z]     @pytest.mark.parametrize('data_gen', [ArrayGen(int_gen)], ids=idfn)

[2025-07-24T11:27:04.910Z]     def test_array_slice_with_negative_length(data_gen, valid_start, negative_length):

[2025-07-24T11:27:04.910Z]         negative_length_gen = IntegerGen(nullable=False, min_val=-25, max_val=-1, special_cases=[])

[2025-07-24T11:27:04.910Z]         # When the list column is all null, the result is also all null regardless of the start and length

[2025-07-24T11:27:04.910Z]         assert_gpu_and_cpu_are_equal_collect(

[2025-07-24T11:27:04.910Z]             lambda spark: three_col_df(spark, array_all_null_gen, array_no_zero_index_gen, negative_length_gen, length=5).selectExpr(

[2025-07-24T11:27:04.910Z]                 f"slice(a, {valid_start}, {negative_length})"))

[2025-07-24T11:27:04.910Z]         error = "The value of parameter(s) `length` in `slice` is invalid: Expects `length` greater than or equal to 0"\

[2025-07-24T11:27:04.910Z]             if is_databricks143_or_later() or is_spark_400_or_later()\

[2025-07-24T11:27:04.910Z]             else 'Unexpected value for length in function slice: length must be greater than or equal to 0.'

[2025-07-24T11:27:04.910Z]         # length can not be negative

[2025-07-24T11:27:04.910Z] >       assert_gpu_and_cpu_error(

[2025-07-24T11:27:04.910Z]             lambda spark: three_col_df(spark, data_gen, array_no_zero_index_gen, negative_length_gen, length=5).selectExpr(

[2025-07-24T11:27:04.911Z]                 f"slice(a, {valid_start}, {negative_length})").collect(),

[2025-07-24T11:27:04.911Z]             conf={},

[2025-07-24T11:27:04.911Z]             error_message=error)

[2025-07-24T11:27:04.911Z] 

[2025-07-24T11:27:04.911Z] ../../src/main/python/array_test.py:397: 

[2025-07-24T11:27:04.911Z] _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

[2025-07-24T11:27:04.911Z] ../../src/main/python/asserts.py:707: in assert_gpu_and_cpu_error

[2025-07-24T11:27:04.911Z]     assert_spark_exception(lambda: with_cpu_session(df_fun, conf), error_message)

[firestarman]

This may be similar as #13037.
The length is 5, so it can produce 5 invalid rows that at least the start or the array is null, then the CPU will not throw out an exception even the length is negative. But GPU probably always checks the length regardless of whether the input array rows and starts are nulls or not.

[firestarman]

FYI: Here is the possible fix at firestarman@9fc3d06