diff --git a/Lib/_strptime.py b/Lib/_strptime.py
--- a/Lib/_strptime.py
+++ b/Lib/_strptime.py
@@ -238,10 +238,7 @@ class TimeRE(dict):
 
         """
         to_convert = sorted(to_convert, key=len, reverse=True)
-        for value in to_convert:
-            if value != '':
-                break
-        else:
+        if all(value == '' for value in to_convert):
             return ''
         regex = '|'.join(re_escape(stuff) for stuff in to_convert)
         regex = '(?P<%s>%s' % (directive, regex)

This looks straightforward.  I believe we normally do not backport this type of change.  The tests for _strptime.py are in test_strptime.py, not test__strptime.py.

Just

    if not any(to_convert):
        return ''

And the conditional return can be done before sorting.

Serhiy, Is there any reason to not combine last two lines in one?

-        regex = '(?P<%s>%s' % (directive, regex)
-        return '%s)' % regex
+        return '(?P<%s>%s)' % (directive, regex)

or to use f-string to then get

+        return f'(?P<{directive}>{regex})'

For me, on Windows, the tests for time, datetime, and _strptime have failures on the master branch, so I don't want to submit a PR just now.

I don't know such reason.

I would not combine f-strings and regular expressions. Curly braces are syntactically meaningful in regular expressions, and regular expression which uses them for substituting look strange.

type: performance
stage: patch review