"Bad" deprecation warning: "Invalid escape sequence" warning incorrectly flags LaTeX

[kmtac]

Bug report

Bug summary

"Bad" deprecation warning: LaTeX $\pm$ raises a DeprecationWarning: Invalid escape sequence \p.
Code for reproduction

import warnings
import matplotlib.pyplot as plt
warnings.simplefiler('always')
plt.text('Silly Error $\pm$ silliness')

Actual outcome

# If applicable, paste the console output here
>>> plt.title('Silly error $\pm$ silliness')
<stdin>:1: DeprecationWarning: invalid escape sequence \p
Text(0.5,1,'Silly error $\\pm$ silliness')

Expected outcome

No error -- This is LaTeX, surrounded by $ $, not an escape sequence. (Note that the plot itself looks fine, with a plus-minus sign.)

Matplotlib version

• Operating system: MacOs X
• Matplotlib version: 2.2.2
• Matplotlib backend: notebook, Qt, MacOSX, Tk
• Python version: 3.6.2
• Jupyter version (if applicable): 4.3.0
• Other libraries: warnings

conda default, conda-forge

[tacaswell]

This is not a Matplotlib warning, this is a warning from cpython

17:11 $ python
Python 3.6.4 (default, Jan  5 2018, 02:35:40) 
[GCC 7.2.1 20171224] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import warnings
>>> warnings.simplefilter('always')
>>> a = '\p'
<stdin>:1: DeprecationWarning: invalid escape sequence \p
>>> 

The fix is to write it as plt.text(r'Silly Error $\pm$ silliness')

[WeatherGod]

Just wondering... is it at all possible to "intercept" these kinds of warnings and add more information to it? Just wondering if it is at all possible.

…

On Mon, Apr 2, 2018 at 5:39 PM, Elliott Sales de Andrade < ***@***.***> wrote: Closed #10945 <#10945>. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#10945 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AARy-Hr_lqGShHti6FzKrPOUAkb_VHJaks5tkpp5gaJpZM4TEEK2> .

[tacaswell]

I suspect the best place to do that would be a PR into upstream cpython.

[WeatherGod]

I was thinking more along the lines of adding on a message that would be relevant in the specific case of matplotlib text labels, like "..., if you were making a LaTeX string, you should make a raw string like so: r'....'" But as I think of it more, I realize that it is impossible because the interpreter spots these problems without any context. Also, the string could be getting passed in via other functions, or being saved as a variable elsewhere. And I doubt that cpython pointing out exactly the invalid escape sequence would be any more useful because that wouldn't solve the confusion that LaTeX has a larger set of escape sequences. I wonder if it is at all possible to detect if a string originated from a raw string literal or not? We could then issue a warning ourselves if we detect a non-raw string.

…

On Mon, Apr 2, 2018 at 10:14 PM, Thomas A Caswell ***@***.***> wrote: I suspect the best place to do that would be a PR into upstream cpython. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#10945 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AARy-AhC0-3Z_l5o5N5-tMnXm1SKUPjvks5tktrugaJpZM4TEEK2> .

[tacaswell]

I'm pretty sure that we can not tell the difference between r'\foo' and '\\foo' (and for now '\foo') just as we can not tell the difference between {} and dict(). The 'r' is an instruction to python on how to interpret the next bit of literal text as it parses it (see https://docs.python.org/3/reference/lexical_analysis.html#string-and-bytes-literals).

I think the best we can do is make sure all of the mathtext / latex docs suggest using 'r'.