characters argument and plt.savefig

[SG-phimeca]

There seem to be a problem when saving figures as pdf with matplotlib.pyplot.savefig and using the characters argument in Waffle: some characters are replaced by others.

list_character = ["🖳", "🐍"]

from pywaffle import Waffle
from matplotlib import pyplot as plt

n_row = 4
n_column = 5

import numpy as np
list_value = np.ones(len(list_character))

fig = plt.figure(
    FigureClass=Waffle,
    rows=1,
    columns=len(list_character),
    values=list_value,
    characters=list_character,
    font_file="~/.fonts/Symbola.ttf",
    font_size=50
    )

plt.savefig("replaced_character.pdf")
plt.close("all")

yields the attached file (two snakes instead of a computer and a snake).
The computer alone is correctly printed. Inverting the order yields two snakes.

replaced_character.pdf

[gyli]

Are you using the latest version, 0.4.1?
I have tested the code above while I can't reproduce this issue.

[SG-phimeca]

Yes, I use 0.4.1. Maybe some other package should be updated?!

[gyli]

Since I can't reproduce the issue, it's kind of tricky to figure out what goes wrong with your result. But it would be great if we can answer the following questions:

• Does it only happen to pdf files, or is it the same result when showing in a pop window or save with other image format?
• Could you check fig._pa to make sure it is the right parameters being passed?
• Could you please find a shareable notebook, such as Azure Notebooks, which can also produce the same result, so that I can take a look and maybe start debugging?

[SG-phimeca]

• yes, it only happens when saving to pdf. Not in interactive mode, nor when saving to scg or png.
• The value associated to 'characters' in fig._pa is correct.
• It might take some time. I'll try.

[gyli]

Interesting. If so, I’m thinking if it’s not the code or package related, but more related to the font or pdf reader. Since you are using a custom font, which might not be included or supported by something.

[SG-phimeca]

Installing pycairo and adding
matplotlib.use('cairo')
before plotting the figure solved the problem.

[gyli]

Good to know that works for you! It does remind me that I have cairo installed on my mac while I didn't switch matplotlib's backend manually.