Optimise import time for string

python · AA-Turner · Apr 8, 2025 · Apr 3, 2025 · Apr 3, 2025 · Apr 3, 2025
commit ec78492ef1ae8db70d6b85c03310fad27af6170d
diff --git a/Lib/string.py b/Lib/string.py
@@ -49,29 +49,20 @@ def capwords(s, sep=None):
 
 
 ####################################################################
-import re as _re
-from collections import ChainMap as _ChainMap
-
+_sentinel_flags = object()
 _sentinel_dict = {}
 
-class Template:
-    """A string class for supporting $-substitutions."""
 
-    delimiter = '$'
-    # r'[a-z]' matches to non-ASCII letters when used with IGNORECASE, but
-    # without the ASCII flag.  We can't add re.ASCII to flags because of
-    # backward compatibility.  So we use the ?a local flag and [a-z] pattern.
-    # See https://bugs.python.org/issue31672
-    idpattern = r'(?a:[_a-z][_a-z0-9]*)'
-    braceidpattern = None
-    flags = _re.IGNORECASE
-
-    def __init_subclass__(cls):
-        super().__init_subclass__()
-        if 'pattern' in cls.__dict__:
+class _TemplatePattern:
+    def __get__(self, instance, cls=None):
+        if cls is None:
+            return self
+        import re
+        if ('pattern' in cls.__dict__
+            and not isinstance(cls.__dict__['pattern'], _TemplatePattern)):
             pattern = cls.pattern
         else:
-            delim = _re.escape(cls.delimiter)
+            delim = re.escape(cls.delimiter)
             id = cls.idpattern
             bid = cls.braceidpattern or cls.idpattern
             pattern = fr"""
@@ -82,7 +73,32 @@ def __init_subclass__(cls):
               (?P<invalid>)             # Other ill-formed delimiter exprs
             )
             """
-        cls.pattern = _re.compile(pattern, cls.flags | _re.VERBOSE)
+        if cls.flags is _sentinel_flags:
+            cls.flags = re.IGNORECASE
+        pattern = re.compile(pattern, cls.flags | re.VERBOSE)
+        # replace this descriptor with the compiled pattern
+        setattr(cls, 'pattern', pattern)
+        return pattern
+
+
+class Template:
+    """A string class for supporting $-substitutions."""
+
+    delimiter = '$'
+    # r'[a-z]' matches to non-ASCII letters when used with IGNORECASE, but
+    # without the ASCII flag.  We can't add re.ASCII to flags because of
+    # backward compatibility.  So we use the ?a local flag and [a-z] pattern.
+    # See https://bugs.python.org/issue31672
+    idpattern = r'(?a:[_a-z][_a-z0-9]*)'
+    braceidpattern = None
+    flags = _sentinel_flags  # default: re.IGNORECASE
+
+    # use a descriptor to be able to defer the import of `re`, for performance
+    pattern = _TemplatePattern()
+
+    def __init_subclass__(cls):
+        super().__init_subclass__()
+        cls.pattern = _TemplatePattern()
 
     def __init__(self, template):
         self.template = template
@@ -105,7 +121,8 @@ def substitute(self, mapping=_sentinel_dict, /, **kws):
         if mapping is _sentinel_dict:
             mapping = kws
         elif kws:
-            mapping = _ChainMap(kws, mapping)
+            from collections import ChainMap
+            mapping = ChainMap(kws, mapping)
         # Helper function for .sub()
         def convert(mo):
             # Check the most common path first.
@@ -124,7 +141,8 @@ def safe_substitute(self, mapping=_sentinel_dict, /, **kws):
         if mapping is _sentinel_dict:
             mapping = kws
         elif kws:
-            mapping = _ChainMap(kws, mapping)
+            from collections import ChainMap
+            mapping = ChainMap(kws, mapping)
         # Helper function for .sub()
         def convert(mo):
             named = mo.group('named') or mo.group('braced')
@@ -170,10 +188,6 @@ def get_identifiers(self):
                     self.pattern)
         return ids
 
-# Initialize Template.pattern.  __init_subclass__() is automatically called
-# only for subclasses, not for the Template class itself.
-Template.__init_subclass__()
-
 
 ########################################################################
 # the Formatter class

diff --git a/Misc/NEWS.d/next/Library/2025-04-03-01-35-02.gh-issue-118761.VQcj70.rst b/Misc/NEWS.d/next/Library/2025-04-03-01-35-02.gh-issue-118761.VQcj70.rst
@@ -0,0 +1,2 @@
+Improve import times by up to 27x for the :mod:`string` module.
+Patch by Adam Turner.
Original file line number	Diff line number	Diff line change
		@@ -0,0 +1,2 @@
		Improve import times by up to 27x for the :mod:`string` module.
		Patch by Adam Turner.