Register metadata in accessor that survives pipes

I was able to achieve this by creating a custom attribute inside the DataFrame and specifying the attribute name in the metadata list:

import pandas as pd
@pd.api.extensions.registerdataframeaccessor("geo")
class GeoAccessor:
    def init(self, pandasobj):
        self.obj = pandasobj
        self.obj.metadata += ["initialshape"]
    def start(self):
        print("Initial shape:", self.obj.shape)
        self.obj.initialshape = self.obj.shape
        return self.obj
    def end(self):
        print("Previous shape:", self.obj.initialshape)
        return self.obj
df = pd.DataFrame({"x": [1, 2, 3], "y": [4, 5, None]})
df.geo.start().dropna(subset=["y"]).geo.end()

returns

Initial shape: (3, 2)
Previous shape: (3, 2)
[The dataframe]