FIX: if bins input to hist is str, treat like no bins

tacaswell · tacaswell · commit 0c8e9d852a28 · 2019-03-31T14:41:54.000-04:00
This change causes the range of all data sets to be computed and passed to numpy (which in turn uses the total range to compute the 'best' bins). The existing code 'latches' the bins from the first data set to use for the rest so this is can still lead to poor binning (if the data sets are widely different). closes #8636
diff --git a/lib/matplotlib/axes/_axes.py b/lib/matplotlib/axes/_axes.py
@@ -6612,7 +6612,9 @@ def hist(self, x, bins=None, range=None, density=None, weights=None,
             bin_range = self.convert_xunits(bin_range)
 
         # Check whether bins or range are given explicitly.
-        binsgiven = np.iterable(bins) or bin_range is not None
+        binsgiven = ((np.iterable(bins) and
+                      not isinstance(bins, str)) or
+                     bin_range is not None)
 
         # We need to do to 'weights' what was done to 'x'
         if weights is not None:
