Fix log contours when using default levels

Author: dstansby

lib/matplotlib/contour.py

@@ -964,7 +964,7 @@ def changed(self):
             label.set_color(self.labelMappable.to_rgba(cv))
         super().changed()
 
-    def _autolev(self, N):
+    def _autolev(self, N, *, using_default):
         """
         Select contour levels to span the data.
 
@@ -981,7 +981,12 @@ def _autolev(self, N):
         """
         if self.locator is None:
             if self.logscale:
-                self.locator = ticker.LogLocator(numticks=N)
+                if using_default:
+                    # Let log locator choose instead of using hard coded value
+                    # set in self._process_contour_level_args()
+                    self.locator = ticker.LogLocator()
+                else:
+                    self.locator = ticker.LogLocator(numticks=N)
             else:
                 self.locator = ticker.MaxNLocator(N + 1, min_n_ticks=1)
 
@@ -1024,8 +1029,9 @@ def _process_contour_level_args(self, args, z_dtype):
                 levels_arg = 7  # Default, hard-wired.
         else:
             levels_arg = self.levels
+
         if isinstance(levels_arg, Integral):
-            self.levels = self._autolev(levels_arg)
+            self.levels = self._autolev(levels_arg, using_default=self.levels is None)
         else:
             self.levels = np.asarray(levels_arg, np.float64)
         if self.filled and len(self.levels) < 2: