Issue #22818: Splitting on a pattern that could match an empty string now

raises a warning.  Patterns that can only match empty strings are now
rejected.

Doc/library/re.rst

@@ -626,17 +626,37 @@ form.
    That way, separator components are always found at the same relative
    indices within the result list.
 
-   Note that *split* will never split a string on an empty pattern match.
+   .. note::
+
+      :func:`split` doesn't currently split a string on an empty pattern match.
       For example:
 
-      >>> re.split('x*', 'foo')
-      ['foo']
-      >>> re.split("(?m)^$", "foo\n\nbar\n")
-      ['foo\n\nbar\n']
+         >>> re.split('x*', 'axbc')
+         ['a', 'bc']
+
+      Even though ``'x*'`` also matches 0 'x' before 'a', between 'b' and 'c',
+      and after 'c', currently these matches are ignored.  The correct behavior
+      (i.e. splitting on empty matches too and returning ``['', 'a', 'b', 'c',
+      '']``) will be implemented in future versions of Python, but since this
+      is a backward incompatible change, a :exc:`FutureWarning` will be raised
+      in the meanwhile.
+
+      Patterns that can only match empty strings currently never split the
+      string.  Since this doesn't match the expected behavior, a
+      :exc:`ValueError` will be raised starting from Python 3.5::
+
+         >>> re.split("^$", "foo\n\nbar\n", flags=re.M)
+         Traceback (most recent call last):
+           File "<stdin>", line 1, in <module>
+           ...
+         ValueError: split() requires a non-empty pattern match.
 
    .. versionchanged:: 3.1
       Added the optional flags argument.
 
+   .. versionchanged:: 3.5
+      Splitting on a pattern that could match an empty string now raises
+      a warning.  Patterns that can only match empty strings are now rejected.
 
 .. function:: findall(pattern, string, flags=0)
 

Doc/whatsnew/3.5.rst

@@ -482,6 +482,13 @@ Changes in the Python API
   simply define :meth:`~importlib.machinery.Loader.create_module` to return
   ``None`` (:issue:`23014`).
 
+* :func:`re.split` always ignored empty pattern matches, so the ``'x*'``
+  pattern worked the same as ``'x+'``, and the ``'\b'`` pattern never worked.
+  Now :func:`re.split` raises a warning if the pattern could match
+  an empty string.  For compatibility use patterns that never match an empty
+  string (e.g. ``'x+'`` instead of ``'x*'``).  Patterns that could only match
+  an empty string (such as ``'\b'``) now raise an error.
+
 Changes in the C API
 --------------------
 

Lib/sre_compile.py

@@ -414,8 +414,11 @@ def _compile_info(code, pattern, flags):
     # this contains min/max pattern width, and an optional literal
     # prefix or a character map
     lo, hi = pattern.getwidth()
+    if hi > MAXCODE:
+        hi = MAXCODE
     if lo == 0:
-        return # not worth it
+        code.extend([INFO, 4, 0, lo, hi])
+        return
     # look for a literal prefix
     prefix = []
     prefixappend = prefix.append
@@ -495,10 +498,7 @@ def _compile_info(code, pattern, flags):
     else:
         emit(MAXCODE)
         prefix = prefix[:MAXCODE]
-    if hi < MAXCODE:
-        emit(hi)
-    else:
-        emit(0)
+    emit(min(hi, MAXCODE))
     # add literal prefix
     if prefix:
         emit(len(prefix)) # length

Lib/test/test_re.py

@@ -251,28 +251,28 @@ class ReTests(unittest.TestCase):
         for string in ":a:b::c", S(":a:b::c"):
             self.assertTypedEqual(re.split(":", string),
                                   ['', 'a', 'b', '', 'c'])
-            self.assertTypedEqual(re.split(":*", string),
+            self.assertTypedEqual(re.split(":+", string),
                                   ['', 'a', 'b', 'c'])
-            self.assertTypedEqual(re.split("(:*)", string),
+            self.assertTypedEqual(re.split("(:+)", string),
                                   ['', ':', 'a', ':', 'b', '::', 'c'])
         for string in (b":a:b::c", B(b":a:b::c"), bytearray(b":a:b::c"),
                        memoryview(b":a:b::c")):
             self.assertTypedEqual(re.split(b":", string),
                                   [b'', b'a', b'b', b'', b'c'])
-            self.assertTypedEqual(re.split(b":*", string),
+            self.assertTypedEqual(re.split(b":+", string),
                                   [b'', b'a', b'b', b'c'])
-            self.assertTypedEqual(re.split(b"(:*)", string),
+            self.assertTypedEqual(re.split(b"(:+)", string),
                                   [b'', b':', b'a', b':', b'b', b'::', b'c'])
         for a, b, c in ("\xe0\xdf\xe7", "\u0430\u0431\u0432",
                         "\U0001d49c\U0001d49e\U0001d4b5"):
             string = ":%s:%s::%s" % (a, b, c)
             self.assertEqual(re.split(":", string), ['', a, b, '', c])
-            self.assertEqual(re.split(":*", string), ['', a, b, c])
-            self.assertEqual(re.split("(:*)", string),
+            self.assertEqual(re.split(":+", string), ['', a, b, c])
+            self.assertEqual(re.split("(:+)", string),
                              ['', ':', a, ':', b, '::', c])
 
-        self.assertEqual(re.split("(?::*)", ":a:b::c"), ['', 'a', 'b', 'c'])
-        self.assertEqual(re.split("(:)*", ":a:b::c"),
+        self.assertEqual(re.split("(?::+)", ":a:b::c"), ['', 'a', 'b', 'c'])
+        self.assertEqual(re.split("(:)+", ":a:b::c"),
                          ['', ':', 'a', ':', 'b', ':', 'c'])
         self.assertEqual(re.split("([b:]+)", ":a:b::c"),
                          ['', ':', 'a', ':b::', 'c'])
@@ -282,11 +282,32 @@ class ReTests(unittest.TestCase):
         self.assertEqual(re.split("(?:b)|(?::+)", ":a:b::c"),
                          ['', 'a', '', '', 'c'])
 
+        for sep, expected in [
+            (':*', ['', 'a', 'b', 'c']),
+            ('(?::*)', ['', 'a', 'b', 'c']),
+            ('(:*)', ['', ':', 'a', ':', 'b', '::', 'c']),
+            ('(:)*', ['', ':', 'a', ':', 'b', ':', 'c']),
+        ]:
+            with self.subTest(sep=sep), self.assertWarns(FutureWarning):
+                self.assertTypedEqual(re.split(sep, ':a:b::c'), expected)
+
+        for sep, expected in [
+            ('', [':a:b::c']),
+            (r'\b', [':a:b::c']),
+            (r'(?=:)', [':a:b::c']),
+            (r'(?<=:)', [':a:b::c']),
+        ]:
+            with self.subTest(sep=sep), self.assertRaises(ValueError):
+                self.assertTypedEqual(re.split(sep, ':a:b::c'), expected)
+
     def test_qualified_re_split(self):
         self.assertEqual(re.split(":", ":a:b::c", maxsplit=2), ['', 'a', 'b::c'])
         self.assertEqual(re.split(':', 'a:b:c:d', maxsplit=2), ['a', 'b', 'c:d'])
         self.assertEqual(re.split("(:)", ":a:b::c", maxsplit=2),
                          ['', ':', 'a', ':', 'b::c'])
+        self.assertEqual(re.split("(:+)", ":a:b::c", maxsplit=2),
+                         ['', ':', 'a', ':', 'b::c'])
+        with self.assertWarns(FutureWarning):
             self.assertEqual(re.split("(:*)", ":a:b::c", maxsplit=2),
                              ['', ':', 'a', ':', 'b::c'])
 

Misc/NEWS

@@ -232,6 +232,10 @@ Core and Builtins
 Library
 -------
 
+- Issue #22818: Splitting on a pattern that could match an empty string now
+  raises a warning.  Patterns that can only match empty strings are now
+  rejected.
+
 - Issue #23099: Closing io.BytesIO with exported buffer is rejected now to
   prevent corrupting exported buffer.
 

Modules/_sre.c

@@ -863,6 +863,19 @@ pattern_split(PatternObject* self, PyObject* args, PyObject* kw)
     if (!string)
         return NULL;
 
+    assert(self->codesize != 0);
+    if (self->code[0] != SRE_OP_INFO || self->code[3] == 0) {
+        if (self->code[0] == SRE_OP_INFO && self->code[4] == 0) {
+            PyErr_SetString(PyExc_ValueError,
+                            "split() requires a non-empty pattern match.");
+            return NULL;
+        }
+        if (PyErr_WarnEx(PyExc_FutureWarning,
+                         "split() requires a non-empty pattern match.",
+                         1) < 0)
+            return NULL;
+    }
+
     string = state_init(&state, self, string, 0, PY_SSIZE_T_MAX);
     if (!string)
         return NULL;