The low level Cython routine for outer_join_indexer uses the wrong size to handle the special case that the right array is empty (see: https://github.com/pydata/pandas/blob/master/pandas/src/generate_code.py#L2134). The fix is simple: if the right array is empty (nright == 0), it should of course loop and copy the left array (for i in range(nleft)).
Due to this bug, the merged index and both indexers are never written when the right array is empty:
In [91]: from pandas.algos import outer_join_indexer_float64
In [92]: outer_join_indexer_float64(array([1.,2.,3.]), array([]))
Out[92]:
(array([ 6.92981993e-310, 2.56210349e-316, 4.27255699e+180]),
array([140261116804248, 140261116804248, 140260374492336]),
array([140261116804232, 140261116804232, 50534992]))
Index.union already preempts this special case separately, so this bug will only affect a handful of people that interact directly with the Cython algos. If desired, I could prepare a PR tomorrow, though it might be overkill for such a small change.
RetroSearch is an open source project built by @garambo | Open a GitHub Issue
Search and Browse the WWW like it's 1997 | Search results from DuckDuckGo
HTML:
3.2
| Encoding:
UTF-8
| Version:
0.7.4