Make TokenList.value a property not an attribute.

The fact that a new value was being computed each time
TokenList.group_tokens() was called caused supra-linear runtime when
token grouping was enabled.

Address by making TokenList.value a dynamically-computed property rather
than a static attribute.

sqlparse/sql.py

@@ -48,6 +48,9 @@ class TokenBase:
     def __init__(self):
         self.parent = None
 
+    def __str__(self):
+        return self.value
+
     # Pending tokenlist __len__ bug fix
     # def __len__(self):
     #     return len(self.value)
@@ -151,9 +154,6 @@ def __init__(self, ttype, value):
         self.is_whitespace = ttype in T.Whitespace
         self.normalized = value.upper() if self.is_keyword else value
 
-    def __str__(self):
-        return self.value
-
     def _get_repr_name(self):
         return str(self.ttype).split('.')[-1]
 
@@ -165,12 +165,11 @@ def flatten(self):
 class TokenList(TokenBase):
     """A group of tokens.
 
-    It has two additional instance attributes:
-    ``value`` is the value of the token and ``tokens`` is a list of child
-    tokens.
+    It has one additional instance attribute, ``tokens``, which is a list of
+    child tokens.
     """
 
-    __slots__ = ('tokens', 'value')
+    __slots__ = 'tokens'
 
     is_group = True
     ttype = None
@@ -180,10 +179,10 @@ class TokenList(TokenBase):
     def __init__(self, tokens=None):
         super().__init__()
         self.tokens = tokens or []
-        self.value = str(self)
         [setattr(token, 'parent', self) for token in self.tokens]
 
-    def __str__(self):
+    @property
+    def value(self):
         return ''.join(token.value for token in self.flatten())
 
     @property
@@ -348,7 +347,6 @@ def group_tokens(self, grp_cls, start, end, include_end=True,
             grp = start
             grp.tokens.extend(subtokens)
             del self.tokens[start_idx + 1:end_idx]
-            grp.value = str(start)
         else:
             subtokens = self.tokens[start_idx:end_idx]
             grp = grp_cls(subtokens)