Add test on Series.getitem failure

[akharche]

No description provided.

[AlexanderKalistratov]

What is the issue? getitem with negative slice is failing?

[kozlov-alexey]

Please use self.subTest context to verify each combination in a separate subtest.
Also I think there's no need to have additional test just for negative slices as they have the same implementation, so it's better to make existing test for int slices, i.e. test_series_getitem_idx_int_slice more generic, e.g:

        slice_args = [0, 2, -2]
        indexes_to_test = {
            'int': [2, 3, 5, 6, 7],
            'float': [2.0, 3, 5, 6, 7],
            'string': ['2', '3', '5', '6', '7']
        }
        for start, end in product(slice_args, slice_args):
            for index in indexes_to_test.values():
                with self.subTest(start=start, stop=end, series_index=index):
                    S = pd.Series([11, 22, 33, 44, 55], index, name='A')
                    result_ref = test_impl(S, start, end)
                    result = jit_impl(S, start, end)
                    pd.testing.assert_series_equal(result, result_ref)

[AlexanderKalistratov]

@kozlov-alexey as far as I get, this is negative tests for negative slices. If it would be combined with positive slices we would need to skip all tests, instead of skiping specific ones.

[kozlov-alexey]

@AlexanderKalistratov What do you mean by negative tests? That an empty slice appear as a result? If so, that's true. Otherwise, I think it should not matter whether slice has negatives or positives from the test perspective (assuming it's a correct slice).

[AlexanderKalistratov]

@kozlov-alexey sounds reasonable. I thought this is some kind of numba limitation (negative slices). But it seems to be our own problem.

[kozlov-alexey]

This would iterate over zipped sequences but until the end of the shortest one is found, so the last string index in indices won't be tested at all.

[kozlov-alexey]

What is the issue? getitem with negative slice is failing?

@akharche @AlexanderKalistratov The issue here is in the empty slice for the string series index only (other combinations seem to pass), i.e. original test fails on the second iteration, when start=-2, end=0 and index is ['2', '3', '5', '6', '7']. And the resulting slice should be:

a = np.asarray([1, 2, 3, 1, 2, 4, 5])
a[-2:0]
array([], dtype=int32)

[kozlov-alexey]

It would probably be better to use the same name (hpat_func) everywhere.

[PokhodenkoSA]

Very old PR. It is better to close it.