Recent hotfix made `sns.relplot` incompatible with latex plotting.

[NotSpecial]

Hi, and first of all thanks for the great library!

I have discovered a problem with a recent hotfix (this line) and latex plotting.

Concretely, the hotfix prepends underscores to all plot variables, e.g. turning x into_x and only changes the labels later.
This causes a problem with the FacetGrid underlying relplot, which calls tight_layout here.

For some reason, this causes matplotlib to render the plot with the axis labels including underscores. I believe it happens to find margins, but I am not sure. In any case, the latex render cannot handle the underscores and crashes.

Affected versions

0.11.2 only; version 0.11.1 (without the hotfix) works fine.

Steps to reproduce

import seaborn as sns
import matplotlib as mpl
import pandas as pd
mpl.rcParams.update({
        'text.usetex': True,
})

testdata = pd.DataFrame({'x': [1, 2], 'y': [2, 3]})
sns.relplot(data=testdata, x='x', y='y');

This raises:

RuntimeError: latex was not able to process the following string:
b'_x'

Possible fixes

Prepend with something more latex friendly instead of underscores, e.g. temp or anything.

[mwaskom]

Ah, vexing. I think \_{var} should work, right?

Unfortunately, displot has a similar problem if you pass a name-less object as a faceting variable.

[NotSpecial]

Yes, \_{var} would work as well 👍

[mwaskom]

@NotSpecial are you still having this issue? I cannot reproduce it on master, using matplotlib 3.5.1. Perhaps matplotlib added something in their latex code to escape leading underscores? This also works fine:

f, ax = plt.subplots()
ax.set_ylabel("_y")

[mwaskom]

I believe this was handled upstream: matplotlib/matplotlib#20706

[NotSpecial]

Great news! I can confirm that it works for me again as well with matplotlib version 3.5.1 and seaborn version 0.11.2. Thanks for checking up on this.