FIX: ValueError when hexbin is run with empty arrays and log scaling. (matplotlib#23944)

* Add test for empty hexbin with log scaling.

* Use guarded autoscale_None for use when vmin/vmax are None.

* Add additional check when auto vmin/vmax are None.

Author: erykoff

lib/matplotlib/axes/_axes.py

@@ -4943,7 +4943,9 @@ def reduce_C_function(C: array) -> float
         # autoscale the norm with current accum values if it hasn't been set
         if norm is not None:
             if norm.vmin is None and norm.vmax is None:
-                norm.autoscale(accum)
+                norm.autoscale_None(accum)
+            norm.vmin = np.ma.masked if norm.vmin is None else norm.vmin
+            norm.vmax = np.ma.masked if norm.vmax is None else norm.vmax
 
         if bins is not None:
             if not np.iterable(bins):

lib/matplotlib/tests/test_axes.py

@@ -901,6 +901,14 @@ def test_hexbin_empty():
     ax.hexbin([], [])
 
 
+@image_comparison(['hexbin_empty.png'], remove_text=True)
+def test_hexbin_log_empty():
+    # From #23922: creating hexbin with log scaling from empty
+    # dataset raises ValueError
+    ax = plt.gca()
+    ax.hexbin([], [], bins='log')
+
+
 def test_hexbin_pickable():
     # From #1973: Test that picking a hexbin collection works
     fig, ax = plt.subplots()