2 tests failures on 32bit: test_searchsorted_sorter and test_sparse_max_row

[yarikoptic]

While building a fresh snapshot (6c692ae) for debians ran into the following consistent one on debian stable and testing (built fine on 64bit, but failed in 32bit userspace):

======================================================================
ERROR: test_searchsorted_sorter (pandas.tests.series.test_analytics.TestSeriesAnalytics)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/build/pandas-0.18.0+git114-g6c692ae/debian/tmp/usr/lib/python2.7/dist-packages/pandas/tests/series/test_analytics.py", line 1331, in test_searchsorted_sorter
    r = s.searchsorted([0, 3], sorter=np.argsort(s))
  File "/build/pandas-0.18.0+git114-g6c692ae/debian/tmp/usr/lib/python2.7/dist-packages/pandas/core/series.py", line 1488, in searchsorted
    side=side, sorter=sorter)
ValueError: could not parse sorter argument

======================================================================
FAIL: test_sparse_max_row (pandas.tests.formats.test_format.TestSeriesFormatting)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/build/pandas-0.18.0+git114-g6c692ae/debian/tmp/usr/lib/python2.7/dist-packages/pandas/tests/formats/test_format.py", line 3724, in test_sparse_max_row
    self.assertEqual(result, exp)
AssertionError: '0    1.0\n1    NaN\n2    NaN\n3    3.0\n4    NaN\ndtype: float64\nBlockIndex\nBlock locations: array([0, 3])\nBlock lengths: array([1, 1])' != '0    1.0\n1    NaN\n2    NaN\n3    3.0\n4    NaN\ndtype: float64\nBlockIndex\nBlock locations: array([0, 3], dtype=int32)\nBlock lengths: array([1, 1], dtype=int32)'

I will replicate in interactive env atm, but thought that may be you have dealt with it already or smth so I wouldn't ;-)

[jreback]

what is your numpy version here? we are still working thru a bunch of numpy issues of numpy > 1.11

[jreback]

cc @sinhrks for the 2nd. though we only test on 32-bit vms very occasionally.

[yarikoptic]

On Tue, 19 Apr 2016, Jeff Reback wrote:

what is your numpy version here? we are still working thru a bunch of
numpy issues of numpy > 1.11

e.g. Debian stable has good old 1.8.2

here is more information about sorter argument

1483
1484 @substitution(klass='Series', value='v')
1485 @appender(base._shared_docs['searchsorted'])
1486 def searchsorted(self, v, side='left', sorter=None):
1487 return self._values.searchsorted(Series(v)._values,
1488 -> side=side, sorter=sorter)
1489
1490 # -------------------------------------------------------------------
1491 # Combination
1492
1493 def append(self, to_append, verify_integrity=False):
(Pdb) p sorter
0 1
1 2
2 0
dtype: int64
(Pdb) p type(sorter)
<class 'pandas.core.series.Series'>

and here is easier to digest (explicit dtype listed) for the 2nd one

(Pdb) p result
'0 1.0\n1 NaN\n2 NaN\n3 3.0\n4 NaN\ndtype: float64\nBlockIndex\nBlock locations: array([0, 3])\nBlock lengths: array([1, 1])'
(Pdb) p exp
'0 1.0\n1 NaN\n2 NaN\n3 3.0\n4 NaN\ndtype: float64\nBlockIndex\nBlock locations: array([0, 3], dtype=int32)\nBlock lengths: array([1, 1], dtype=int32)'

[sinhrks]

Is 2nd case has been fixed by #12936? (thanks, @jreback)

[jreback]

@yarikoptic master should be clean now on 32bit platforms.