Implement Series.sum() in new style

[PokhodenkoSA]

Implemented Series.sum() via numpy.nansum().
Is it correct solution or should I implement it in another way?

[shssf]

Do we have a tests for this functionality? Did you try them with HPAT_CONFIG_PIPELINE_HPAT=0 to make sure Numba could work with it?
Overall approach is good.

[PokhodenkoSA]

With HPAT_CONFIG_PIPELINE_HPAT=0 the old test hpat.tests.test_series.TestSeries.test_series_sum2 do not work:

No definition for lowering <built-in function add>(series(float64, array(float64, 1d, C), none, False), series(float64, array(float64, 1d, C), none, False)) -> series(float64, array(float64, 1d, C), none, False)

File "hpat\tests\test_series.py", line 859:
        def test_impl(S):
            return (S + S).sum()
            ^

[1] During: lowering "$0.3 = S + S" at c:\Users\spokhode\Documents\hpat-workspace\hpat\hpat\tests\test_series.py (859)

As I can understand add operator for Series is not implemented yet. @shssf should I fix it in this PR or in new one?

[shssf]

@PokhodenkoSA I see the test (return (S + S).sum()) is testing two diffrent things: function sum() and operator+
This test should be disabled until we implement all required functionality. Example in PR #189 (@unittest.skipIf(hpat.config.config_pipeline_hpat_default can be used)
Please create new tests for particular sum()

Operators should be implemented separately (in other PRs).

[shssf]

please resolve conflicts

[PokhodenkoSA]

@shssf, I need help.

I have made all modifications we usually do for disabling old approach for sum:

1. Implement @overload_method in hpat/datatypes/hpat_pandas_series_functions.py
2. Removed sum from list in _run_call_series() in hpat/hiframes/hiframes_typed.py
3. Disable adding resolve_sum to SeriesAttribute in hpat/hiframes/pd_series_ext.py
4. Commented/uncommented sum in hpat/hiframes/series_kernels.py:

series_replace_funcs = {
    # 'sum': _column_sum_impl_basic,

Moreover, I have fixed call for hpat.hiframes.pd_series_ext.SeriesAttribute.resolve_sum() in hpat/hiframes/pd_dataframe_ext.py (SeriesAttribute has no resolve_sum, see point 3 above).

Finally I got this errors:

ERROR: test_str_contains_noregex (hpat.tests.test_hiframes.TestHiFrames)
  File "...\AppData\Local\Continuum\anaconda3\envs\hpat-sum-env\lib\site-packages\numba\datamodel\models.py", line 696, in getter
    raise TypeError("expecting {0} but got {1}".format(*args))
TypeError: expecting {i8*, i8*, i64, i64, i8*, [1 x i64], [1 x i64]} but got {{i8*, i8*, i64, i64, i8*, [1 x i64], [1 x i64]}, i8*, i8*}

ERROR: test_str_contains_regex (hpat.tests.test_hiframes.TestHiFrames)
  File "C:\Users\spokhode\AppData\Local\Continuum\anaconda3\envs\hpat-sum-env\lib\site-packages\numba\datamodel\models.py", line 696, in getter
    raise TypeError("expecting {0} but got {1}".format(*args))
TypeError: expecting {i8*, i8*, i64, i64, i8*, [1 x i64], [1 x i64]} but got {{i8*, i8*, i64, i64, i8*, [1 x i64], [1 x i64]}, i8*, i8*}

And many errors like that in different test suites:

======================================================================
FAIL: test_column_distribution (hpat.tests.test_hiframes.TestHiFrames)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\Users\spokhode\Documents\hpat-workspace\sum-workspace\hpat\hpat\tests\test_hiframes.py", line 161, in test_column_distribution
    self.assertEqual(count_array_REPs(), 0)
AssertionError: 5 != 0

I have no idea what to do else.
@shssf, Have you any suggestion what to try or what could be wrong?

[shssf]

@PokhodenkoSA I will take a look

[PokhodenkoSA]

I have found solution (thanks to @kozlov-alexey for pointing to this).
I have returned sum to list in _run_call_series() in hpat/hiframes/hiframes_typed.py (step 2)
and uncommented 'sum' in hpat/hiframes/series_kernels.py (step 4).
Also I have skipped test for skipna because old style does not allow parameters for sum().

[shssf]

@PokhodenkoSA I started looking for this PR today and didn't understand - everything worked well.
later I found you made a changes which make this PR working :-) It is good!
Thank you for squashing commits 👍

[PokhodenkoSA]

@shssf Thank you for merging it.
BTW, GitHub can squash commits before merge https://github.blog/2016-04-01-squash-your-commits/.

@fschlimb could you please check that this feature is enabled in settings of the repository?