Referencing hierarchical columns gives error in seaborn 0.11

[Maltimore]

I updated to seaborn 0.11 and get an error with code that did not even give a warning with seaborn 0.10.

I have a pandas DataFrame with hierarchical columns. That means that I reference a column like so:

df[('0th_order_col_name', '1st_order_col_name')]

When I try to specify such a hierarchical column label in, for instance, a scatterplot, seaborn now gives me an error because it tries to interpret this hierarchical column label as data.

sns.scatterplot(x=('0th_order_col_name', '1st_order_col_name_FOO'), y=('0th_order_col_name', '1st_order_col_name_BAR'), data=df)

The error message is

ValueError: Length of tuple vectors must match length of `data` when both are used, but `data` has length 5 and the vector passed to `x` has length 2.

The change was introduced in #2205

[mwaskom]

I don't think this was intentionally supported to begin with, and I'm a little surprised that it did.

[Maltimore]

Are you planning to support this in the future then, or should I close this issue?

[mwaskom]

Hierarchical columns add a lot of complexity. They're not "tidy", strictly speaking. My inclination would be to say that they're not supported, because it was never claimed that they were. But it does look like they worked, accidentally, in more places than I realized. So I can look into how disruptive it would be to handle them in the refactored code.