avoid quadratic trailing-char search in _extract_possible_number

python/phonenumbers/phonenumberutil.py

@@ -248,7 +248,10 @@
 # which nets down to: >=1 non-Number, non-Letter, non-# characters at string end
 # In Python Unicode regexp mode '(?u)', the class '[^#\w]' will match anything
 # that is not # and is not alphanumeric and is not underscore.
-_UNWANTED_END_CHARS = u(r"(?u)(?:_|[^#\w])+$")
+# The negative lookbehind only lets a match start at the beginning of a run
+# of these characters; without it, search() retries from every position in a
+# run that is not at the end of the string, which is quadratic in its length.
+_UNWANTED_END_CHARS = u(r"(?u)(?<!_|[^#\w])(?:_|[^#\w])+$")
 _UNWANTED_END_CHAR_PATTERN = re.compile(_UNWANTED_END_CHARS)
 
 # We use this pattern to check if the phone number has at least three letters

python/tests/shortnumberinfotest.py

@@ -17,6 +17,7 @@
 # limitations under the License.
 
 import sys
+import time
 
 import phonenumbers
 from phonenumbers import connects_to_emergency_number, is_emergency_number, ShortNumberCost
@@ -269,6 +270,15 @@ def testIsEmergencyNumberLongNumber_BR(self):
         self.assertFalse(is_emergency_number("1900", "BR"))
         self.assertFalse(is_emergency_number("9996", "BR"))
 
+    def testEmergencyNumberLongSeparatorRun(self):
+        # Python version extra test: a long run of separators that isn't at
+        # the end of the input used to take quadratic time to strip.
+        number = "911" + " " * 20000 + "1"
+        start = time.time()
+        self.assertFalse(is_emergency_number(number, "US"))
+        self.assertTrue(connects_to_emergency_number(number, "US"))
+        self.assertTrue(time.time() - start < 1)
+
     def testIsEmergencyNumber_AO(self):
         # Angola doesn't have any metadata for emergency numbers in the test metadata.
         self.assertFalse(is_emergency_number("911", "AO"))