REF: use fused types for join_helper #23171
Conversation
|
Hello @jbrockmendel! Thanks for submitting the PR.
|
|
I'm seeing this same Travis failure in several unrelated PRs. |
|
I haven't made any progress debugging it though. |
Codecov Report
@@ Coverage Diff @@
## master #23171 +/- ##
==========================================
- Coverage 92.19% 92.19% -0.01%
==========================================
Files 169 169
Lines 50959 50956 -3
==========================================
- Hits 46980 46977 -3
Misses 3979 3979
Continue to review full report at Codecov.
|
jreback
added
Refactor
| left_join_indexer_unique_uint64 = left_join_indexer_unique["uint64_t"] | ||
|
|
||
|
|
||
| # @cython.wraparound(False) |
There was a problem hiding this comment.
Maintaining the status quo.
pandas/core/indexes/numeric.py
Outdated
| _default_dtype = np.float64 | ||
|
|
||
| # Cython methods; see github.com/cython/cython/issues/2647 | ||
| # for why we need to wrap these instead of making them class attributes | ||
| def _left_indexer_unique(self, left, right): |
There was a problem hiding this comment.
we could think about doing generic dispatch in the base class,
e.g.
def _left_indexer(self, left, right):
return getattr(libjoin, 'left_join_indexer_{}.format(self._join_type))(left, right)
might be slightly less readable, but less code overall in the subclasses.
There was a problem hiding this comment.
Actually we can do one better. By just calling left_join_indexer cython will dispatch to the appropriate version on its own. (according to the docs; I haven't tried this myself)
|
thanks! |
Related but orthogonal to #23022.