[pypy-issue] [issue1022] dict.update significantly slower than series of dict.__setitem__

[Xavier Morel]

New submission from Xavier Morel <bugs.pypy.org at masklinn.net>:

In CPython, as the number of keys grow dict.update settles at roughly half the 
speed of an 
equivalent bunch of dict[key]=value sequence:

    > python2.7 -m timeit -s 'd = {}' 'd["foo"] = 3' 'd["bar"] = 4' 'd["baz"] = 5' 
'd["qux"] = 6' 'd["quux"] = 7'
    1000000 loops, best of 3: 0.395 usec per loop
    > python2.7 -m timeit -s 'd = {}' 'd.update(foo=3, bar=4, baz=5, qux=6, 
quux=7)'
    1000000 loops, best of 3: 0.657 usec per loop

In Pypy, on the other hand, the difference in speed is on the order of 20x, even 
while 
setting 5 keys and dict.update is ~30% the speed of the equivalent call in CPython 
at 
almost 2µsec/loop to CPython's 0.66µsec/loop, while pypy's __setitem__ is more 
than 4 times 
faster than CPython:

    > pypy-c -m timeit -s 'd = {}' 'd["foo"] = 3' 'd["bar"] = 4' 'd["baz"] = 5' 
'd["qux"] 
= 6' 'd["quux"] = 7'
    10000000 loops, best of 3: 0.0877 usec per loop
    > pypy-c -m timeit -s 'd = {}' 'd.update(foo=3, bar=4, baz=5, qux=6, quux=7)'
    100000 loops, best of 3: 1.95 usec per loop

that kind-of bothers me, as I tend to use dict.update over long sequences of 
setitem for 
readability purpose. Now of course half a µsec per call is not a big difference in 
absolute, but still…

Versions information:
* OSX 10.6.8
* CPython 2.7.2 (Macports)
* Pypy 1.7.0 Python 2.7.1 (?, Nov 24 2011, 10:57:50) (Macports)

----------
messages: 3814
nosy: masklinn, pypy-issue
priority: performance bug
status: unread
title: dict.update significantly slower than series of dict.__setitem__

________________________________________
PyPy bug tracker <tracker at bugs.pypy.org>
<https://bugs.pypy.org/issue1022>
________________________________________