gh-111259: Optimize complementary character sets in RE (GH-120742)

Patterns like "[\s\S]" or "\s|\S" which match any character are now compiled
to the same effective code as a dot with the DOTALL modifier ("(?s:.)").

Lib/re/_compiler.py

@@ -28,6 +28,8 @@
     POSSESSIVE_REPEAT: (POSSESSIVE_REPEAT, SUCCESS, POSSESSIVE_REPEAT_ONE),
 }
 
+_CHARSET_ALL = [(NEGATE, None)]
+
 def _combine_flags(flags, add_flags, del_flags,
                    TYPE_FLAGS=_parser.TYPE_FLAGS):
     if add_flags & TYPE_FLAGS:
@@ -84,17 +86,22 @@ def _compile(code, pattern, flags):
                     code[skip] = _len(code) - skip
         elif op is IN:
             charset, hascased = _optimize_charset(av, iscased, tolower, fixes)
-            if flags & SRE_FLAG_IGNORECASE and flags & SRE_FLAG_LOCALE:
-                emit(IN_LOC_IGNORE)
-            elif not hascased:
-                emit(IN)
-            elif not fixes:  # ascii
-                emit(IN_IGNORE)
+            if not charset:
+                emit(FAILURE)
+            elif charset == _CHARSET_ALL:
+                emit(ANY_ALL)
             else:
-                emit(IN_UNI_IGNORE)
-            skip = _len(code); emit(0)
-            _compile_charset(charset, flags, code)
-            code[skip] = _len(code) - skip
+                if flags & SRE_FLAG_IGNORECASE and flags & SRE_FLAG_LOCALE:
+                    emit(IN_LOC_IGNORE)
+                elif not hascased:
+                    emit(IN)
+                elif not fixes:  # ascii
+                    emit(IN_IGNORE)
+                else:
+                    emit(IN_UNI_IGNORE)
+                skip = _len(code); emit(0)
+                _compile_charset(charset, flags, code)
+                code[skip] = _len(code) - skip
         elif op is ANY:
             if flags & SRE_FLAG_DOTALL:
                 emit(ANY_ALL)
@@ -277,6 +284,10 @@ def _optimize_charset(charset, iscased=None, fixup=None, fixes=None):
                             charmap[i] = 1
                 elif op is NEGATE:
                     out.append((op, av))
+                elif op is CATEGORY and tail and (CATEGORY, CH_NEGATE[av]) in tail:
+                    # Optimize [\s\S] etc.
+                    out = [] if out else _CHARSET_ALL
+                    return out, False
                 else:
                     tail.append((op, av))
             except IndexError:
@@ -519,13 +530,18 @@ def _compile_info(code, pattern, flags):
     # look for a literal prefix
     prefix = []
     prefix_skip = 0
-    charset = [] # not used
+    charset = None # not used
     if not (flags & SRE_FLAG_IGNORECASE and flags & SRE_FLAG_LOCALE):
         # look for literal prefix
         prefix, prefix_skip, got_all = _get_literal_prefix(pattern, flags)
         # if no prefix, look for charset prefix
         if not prefix:
             charset = _get_charset_prefix(pattern, flags)
+            if charset:
+                charset, hascased = _optimize_charset(charset)
+                assert not hascased
+                if charset == _CHARSET_ALL:
+                    charset = None
 ##     if prefix:
 ##         print("*** PREFIX", prefix, prefix_skip)
 ##     if charset:
@@ -560,8 +576,6 @@ def _compile_info(code, pattern, flags):
         # generate overlap table
         code.extend(_generate_overlap_table(prefix))
     elif charset:
-        charset, hascased = _optimize_charset(charset)
-        assert not hascased
         _compile_charset(charset, flags, code)
     code[skip] = len(code) - skip
 

Lib/re/_constants.py

@@ -206,6 +206,8 @@ def _makecodes(*names):
     CATEGORY_NOT_LINEBREAK: CATEGORY_UNI_NOT_LINEBREAK
 }
 
+CH_NEGATE = dict(zip(CHCODES[::2] + CHCODES[1::2], CHCODES[1::2] + CHCODES[::2]))
+
 # flags
 SRE_FLAG_IGNORECASE = 2 # case insensitive
 SRE_FLAG_LOCALE = 4 # honour system locale

Lib/test/test_re.py

@@ -2473,6 +2473,24 @@ def test_regression_gh94675(self):
     def test_fail(self):
         self.assertEqual(re.search(r'12(?!)|3', '123')[0], '3')
 
+    def test_character_set_any(self):
+        # The union of complementary character sets mathes any character
+        # and is equivalent to "(?s:.)".
+        s = '1x\n'
+        for p in r'[\s\S]', r'[\d\D]', r'[\w\W]', r'[\S\s]', r'\s|\S':
+            with self.subTest(pattern=p):
+                self.assertEqual(re.findall(p, s), list(s))
+                self.assertEqual(re.fullmatch('(?:' + p + ')+', s).group(), s)
+
+    def test_character_set_none(self):
+        # Negation of the union of complementary character sets does not match
+        # any character.
+        s = '1x\n'
+        for p in r'[^\s\S]', r'[^\d\D]', r'[^\w\W]', r'[^\S\s]':
+            with self.subTest(pattern=p):
+                self.assertIsNone(re.search(p, s))
+                self.assertIsNone(re.search('(?s:.)' + p, s))
+
 
 def get_debug_out(pat):
     with captured_stdout() as out:

Misc/NEWS.d/next/Library/2024-06-19-13-20-01.gh-issue-111259.Wki5PV.rst

@@ -0,0 +1,3 @@
+:mod:`re` now handles patterns like ``"[\s\S]"`` or ``"\s|\S"`` which match
+any character as effectively as a dot with the ``DOTALL`` modifier
+(``"(?s:.)"``).