Black: Regression in 20.8b1: "skip-string-normalisation" is ignored for docstrings

[exhuma]

Describe the bug

As of version 20.8b1, the "skip string normalisation" setting is ignored, but only for docstrings

To Reproduce Steps to reproduce the behavior:

Example code

class Hello:
    '''
    Wrong Strings
    '''


def foo():
    '''
    wrong-string
    '''

1. Run _Black_ on it with these arguments '....'

black --check --diff -S foo.py

1. See error

--- foo.py      2020-08-27 06:15:44.100921 +0000
+++ foo.py      2020-08-27 06:16:07.598050 +0000
@@ -1,10 +1,10 @@
 class Hello:
-    '''
+    """
     Wrong Strings
-    '''
+    """


 def foo():
-    '''
+    """
     wrong-string
-    '''
+    """
would reformat foo.py
Oh no! 💥 💔 💥
1 file would be reformatted.

Expected behavior

While it is true that single-quoted strings violate PEP-8 via PEP-257 I would still prefer if that the behaviour of black would not change in that respect. In 19.10b0 it was still honouring the "skip-string-normalisation" setting, even for docstrings. This has now changed.

Environment (please complete the following information):

• Version: 20.8b1
• OS and Python version: Linux/Python 3.7.0

Does this bug also happen on master?

yes

Additional context

Our team has a fairly large number of Python projects. I just checked my own listing and I have over 170 projects myself, so I assume our team will have easily over 200 projects. Most small libraries, but some are fairly large code-bases. A lot of code conventions are historic and we're switching to black at the moment. The skip-string-normalisation was/is a big helping hand in that migration, as it would otherwise just simply cause too many diffs & thus conflicts.

This option allowed us to have a fairly painless migration to black as we were quite close to PEP-8 already. But the string-normalisation hamstringed us a bit. And now with 20.8b1 it's back. And this morning, coming to the office we were showered with CI-emails of our daily runs with pipeline failures 😦

[TylerYep]

I believe this has already been fixed in this commit

[exhuma]

This is good to know. Thanks for the info.

In case anyone else comes across this, I have fixed our pipelines for now by blacklisting both 20.8b0 and 20.8b1 using the following command:

pip install --pre "black != 20.8b0, != 20.8b1"

Instead of pinning this using either < 20.8b0 or ==19.10b0, the blacklisting syntax will ensure we get any upcoming release of black in our daily pipelines.

Considering that this seems to be handled in the commit mentioned above, I think it's safe enough to close this issue with the workaround I just laid out.

If it breaks again in a newer release I will notice fairly quickly and update this ticket if need be.

[hugovk]

To avoid dozens of CI failure emails, I'd suggest pinning formatters and linters like Black and Flake8 to a single, specific version on CI and locally. And then making explicit upgrades from time to time.

[ambv]

Closing as duplicate of #1634.