Michael Waskom

To be clear, in the seaborn usecase, it is 100% intentional to be sending in string-typed numbers.

Right, it would be easier for me overall if there were just no feedback here, though I am sympathetic about this being a common gotcha. But whereas with `warnings.catch_warnings`/`warnings.simplefilter` there...

As a maybe orthogonal conversation, maybe `axis.update_units` could have a "verbosity" parameter that sets some state to prevent messages (logging or warning) of this kind. Would need some additional thinking...

This issue is about the objects interface (where what you want actually is possible), so that comment is off topic. You may want to read through #2231

> I’m tempted to deprecate labels-only completely. Please do this! People get themselves into so much trouble trying to customize seaborn legends this way, and I don't really know what...

The official seaborn line is that you should pass in the values that you want to appear in the legend. That’s not ideal and I would prefer it if the...

Yeah I guess, but for seaborn there's not really much benefit to that over calling `rename` on the dataframe you're sending into the plot. For matplotlib usecases I think it's...

Yeah this is a rough edge. The easiest thing to do would be to delegate this to matplotlib, but their auto layout engines can't handle a figure legend. There's some...

If you look at the traceback, this is coming directly out of matplotlib: ```python-traceback --------------------------------------------------------------------------- TypeError Traceback (most recent call last) Cell In [25], line 1 ----> 1 plt.pcolormesh(df.astype('Int64')) File...

(oops sorry the traceback I copied is what you get with `plt.pcolormesh`, but it's the same thing you get with `heatmap` just missing a frame or two within seaborn.)