Relax conditions for warning on updating converters

ksunden · ksunden · commit 530ac3d77533 · 2024-11-18T17:01:48.000-06:00
Allow sub/superclass without warning.

This was motivated by Pandas, as they have several instances of the same class in the registry.
We also had this internally, though resolved that by using the same instance for multiple registry
entries rather than by relaxing the warning. This becomes impractical when dealing with downstream.

It may have been sufficient to allow only identical types, but to be safe, decided to allow both
sub- and superclass relationships.
The purpose of the warning was that it gets overwritten without heeding prior data provided which
would break in unexpected ways.
The most common case for this is when one converter class operates on multiple data types, but
separate instances of the class are in the registry. So while the previous data is compatible the
equality check still failed. Requiring downstream to implement equality checks seems impractical.

Additionally used both `is` and `==` for the earlier short circuit, which should in most cases be
identical operations, but just in case.
diff --git a/lib/matplotlib/axis.py b/lib/matplotlib/axis.py
@@ -1842,11 +1842,15 @@ def set_converter(self, converter):
         self._converter_is_explicit = True
 
     def _set_converter(self, converter):
-        if self._converter == converter:
+        if self._converter is converter or self._converter == converter:
             return
         if self._converter_is_explicit:
             raise RuntimeError("Axis already has an explicit converter set")
-        elif self._converter is not None:
+        elif (
+            self._converter is not None and
+            not isinstance(converter, type(self._converter)) and
+            not isinstance(self._converter, type(converter))
+        ):
             _api.warn_external(
                 "This axis already has a converter set and "
                 "is updating to a potentially incompatible converter"
