Backport PR matplotlib#29897: BUG: ensure that errorbar does not error on masked negative errors.

Author: QuLogic

lib/matplotlib/axes/_axes.py

@@ -3813,9 +3813,12 @@ def apply_mask(arrays, mask):
                     f"'{dep_axis}err' must not contain None. "
                     "Use NaN if you want to skip a value.")
 
-            res = np.zeros(err.shape, dtype=bool)  # Default in case of nan
-            if np.any(np.less(err, -err, out=res, where=(err == err))):
-                # like err<0, but also works for timedelta and nan.
+            # Raise if any errors are negative, but not if they are nan.
+            # To avoid nan comparisons (which lead to warnings on some
+            # platforms), we select with `err==err` (which is False for nan).
+            # Also, since datetime.timedelta cannot be compared with 0,
+            # we compare with the negative error instead.
+            if np.any((check := err[err == err]) < -check):
                 raise ValueError(
                     f"'{dep_axis}err' must not contain negative values")
             # This is like

lib/matplotlib/tests/test_axes.py

@@ -4438,10 +4438,23 @@ def test_errorbar_nan(fig_test, fig_ref):
     xs = range(5)
     ys = np.array([1, 2, np.nan, np.nan, 3])
     es = np.array([4, 5, np.nan, np.nan, 6])
-    ax.errorbar(xs, ys, es)
+    ax.errorbar(xs, ys, yerr=es)
     ax = fig_ref.add_subplot()
-    ax.errorbar([0, 1], [1, 2], [4, 5])
-    ax.errorbar([4], [3], [6], fmt="C0")
+    ax.errorbar([0, 1], [1, 2], yerr=[4, 5])
+    ax.errorbar([4], [3], yerr=[6], fmt="C0")
+
+
+@check_figures_equal()
+def test_errorbar_masked_negative(fig_test, fig_ref):
+    ax = fig_test.add_subplot()
+    xs = range(5)
+    mask = np.array([False, False, True, True, False])
+    ys = np.ma.array([1, 2, 2, 2, 3], mask=mask)
+    es = np.ma.array([4, 5, -1, -10, 6], mask=mask)
+    ax.errorbar(xs, ys, yerr=es)
+    ax = fig_ref.add_subplot()
+    ax.errorbar([0, 1], [1, 2], yerr=[4, 5])
+    ax.errorbar([4], [3], yerr=[6], fmt="C0")
 
 
 @image_comparison(['hist_stacked_stepfilled', 'hist_stacked_stepfilled'])