Raymond Hettinger <[email protected]> added the comment:
Some thoughts:
* Other than set operations, most of the pure python code in the dict view ABCs
are fast pass throughs. There is no point in rewriting these in C:
def __contains__(self, key):
return key in self._mapping
def __iter__(self):
yield from self._mapping
def __len__(self):
return len(self._mapping)
* For the set operations, the pure python code has only a simple for-loop and
contains test. There isn't much fat to be cut:
def __ge__(self, other):
if not isinstance(other, Set):
return NotImplemented
if len(self) < len(other):
return False
for elem in other:
if elem not in self:
return False
return True
* Possibly the eval-loop overhead can be eliminated with Python by using
itertools:
def __ge__(self, other):
if not isinstance(other, Set):
return NotImplemented
if len(self) < len(other):
return False
for elem in filterfalse(self.__contains__, other):
return False
return True
* That leaves the question of why the dict views are so much faster (presuming
that the posted timings are representative). I haven't looked in detail, but
the first candidate that comes to mind is that dictviews_to_set() has a fast
path for exact dicts. That lets it bypass the mapping proxy and exploit the
fast path in PySet_New for exact dicts. That fast path reuses the stored hash
values and exploits knowing that the input has no duplicates. Off-hand, I
don't see how that can generalize to the arbitrary mappings supports by the
view ABCs.
* Summary: A first look, it doesn't seem like a C rewrite of the view ABCs
would bear fruit.
----------
_______________________________________
Python tracker <[email protected]>
<https://bugs.python.org/issue46713>
_______________________________________
_______________________________________________
Python-bugs-list mailing list
Unsubscribe:
https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
