gh-87888: Make FileCookieJar use latin-1 instead of locale encoding

* Byte transparent reading and writing.
* Consistent with WSGI.
* Avoid UnicodeDecodeError when reading comments.

Author: methane

Lib/http/cookiejar.py

@@ -1798,7 +1798,9 @@ def load(self, filename=None, ignore_discard=False, ignore_expires=False):
             if self.filename is not None: filename = self.filename
             else: raise ValueError(MISSING_FILENAME_TEXT)
 
-        with open(filename) as f:
+        # We use latin-1 here because WSGI uses latin-1 for HTTP headers too.
+        # See gh-87888 for more info.
+        with open(filename, encoding="latin1") as f:
             self._really_load(f, filename, ignore_discard, ignore_expires)
 
     def revert(self, filename=None,
@@ -1890,7 +1892,7 @@ def save(self, filename=None, ignore_discard=False, ignore_expires=False):
             if self.filename is not None: filename = self.filename
             else: raise ValueError(MISSING_FILENAME_TEXT)
 
-        with open(filename, "w") as f:
+        with open(filename, "w", encoding="latin1") as f:
             # There really isn't an LWP Cookies 2.0 format, but this indicates
             # that there is extra information in here (domain_dot and
             # port_spec) while still being compatible with libwww-perl, I hope.
@@ -2086,7 +2088,7 @@ def save(self, filename=None, ignore_discard=False, ignore_expires=False):
             if self.filename is not None: filename = self.filename
             else: raise ValueError(MISSING_FILENAME_TEXT)
 
-        with open(filename, "w") as f:
+        with open(filename, "w", encoding="latin1") as f:
             f.write(NETSCAPE_HEADER_TEXT)
             now = time.time()
             for cookie in self:

Misc/NEWS.d/next/Library/2022-05-06-20-32-41.gh-issue-87888.2_R3zS.rst

@@ -0,0 +1,2 @@
+Changed encoding used by :class:`http.cookiejar.FileCookieJar` and its
+subclasses from locale encoding to "latin-1".