Simplify test for negative xerr/yerr.

1) Move the test into x/y loop, to write it just once.
2) Change the test to directly also work for timedelta, rather than
   having to manually test it.  (In practice, the relevant question is
   whether the "low" end of the errorbar is indeed below the "high" end
   of the errorbar.)

Author: anntzer

lib/matplotlib/axes/_axes.py

@@ -3,7 +3,6 @@
 import logging
 import math
 from numbers import Integral, Number
-from datetime import timedelta
 
 import numpy as np
 from numpy import ma
@@ -3291,19 +3290,6 @@ def errorbar(self, x, y, yerr=None, xerr=None,
         if len(x) != len(y):
             raise ValueError("'x' and 'y' must have the same size")
 
-        def has_negative_values(array):
-            if array is None:
-                return False
-            try:
-                return np.any(array < 0)
-            except TypeError:  # if array contains 'datetime.timedelta' types
-                return np.any(array < timedelta(0))
-
-        if has_negative_values(xerr):
-            raise ValueError("'xerr' must not contain negative values")
-        if has_negative_values(yerr):
-            raise ValueError("'yerr' must not contain negative values")
-
         if isinstance(errorevery, Integral):
             errorevery = (0, errorevery)
         if isinstance(errorevery, tuple):
@@ -3426,6 +3412,9 @@ def apply_mask(arrays, mask): return [array[mask] for array in arrays]
                     f"'{dep_axis}err' (shape: {np.shape(err)}) must be a "
                     f"scalar or a 1D or (2, n) array-like whose shape matches "
                     f"'{dep_axis}' (shape: {np.shape(dep)})") from None
+            if np.any(err < -err):  # like err<0, but also works for timedelta.
+                raise ValueError(
+                    f"'{dep_axis}err' must not contain negative values")
             # This is like
             #     elow, ehigh = np.broadcast_to(...)
             #     return dep - elow * ~lolims, dep + ehigh * ~uplims