[pypy-commit] pypy keys_with_hash: Identified a difference of performance with CPython: at least in

[arigo]

Author: Armin Rigo <arigo at tunes.org>
Branch: keys_with_hash
Changeset: r79325:8ca42034ab3f
Date: 2015-08-31 16:47 +0200
http://bitbucket.org/pypy/pypy/changeset/8ca42034ab3f/

Log:	Identified a difference of performance with CPython: at least in
	RPython code, there are places where we should reuse the hash of a
	dict key instead of recomputing it

diff --git a/pypy/objspace/std/setobject.py b/pypy/objspace/std/setobject.py
--- a/pypy/objspace/std/setobject.py
+++ b/pypy/objspace/std/setobject.py
@@ -965,6 +965,9 @@
         other_dict = self.unerase(w_other.sstorage)
         result_dict = self.get_empty_dict()
         for key in iterator:
+            # xxx performance issue when compared to CPython: the next
+            # two lines will recompute twice the hash of 'key', whereas
+            # CPython reuses the hash from 'iterator' in both cases.
             if key not in other_dict:
                 result_dict[key] = None
         return self.erase(result_dict)