BUG: Ignore division by 0 when merging empty dataframes (pandas-dev#17776)

Author: yeemey

doc/source/whatsnew/v0.21.0.txt

@@ -1002,6 +1002,7 @@ Reshaping
 - Bug in :func:`concat` where order of result index was unpredictable if it contained non-comparable elements (:issue:`17344`)
 - Fixes regression when sorting by multiple columns on a ``datetime64`` dtype ``Series`` with ``NaT`` values (:issue:`16836`)
 - Bug in :fun:`pivot_table` where the result's columns did not preserve the categorical dtype of ``columns`` when ``dropna`` was ``False`` (:issue:`17842`)
+- Bug in merging two empty dataframes, which raised a division by zero warning (:issue:`17776`)
 
 Numeric
 ^^^^^^^

pandas/core/reshape/merge.py

@@ -1514,6 +1514,7 @@ def _sort_labels(uniques, left, right):
 
 
 def _get_join_keys(llab, rlab, shape, sort):
+    np.seterr(divide='ignore')
 
     # how many levels can be done without overflow
     pred = lambda i: not is_int64_overflow_possible(shape[:i])
@@ -1528,6 +1529,7 @@ def _get_join_keys(llab, rlab, shape, sort):
         stride //= shape[i]
         lkey += llab[i] * stride
         rkey += rlab[i] * stride
+    np.seterr(divide='warn')
 
     if nlev == len(shape):  # all done!
         return lkey, rkey