[Python-Dev] Comparison speed

[Barry A. Warsaw]

>>>>> "JH" == Jeremy Hylton <jeremy@digicool.com> writes:

    JH> I only learned recently that isinstance() can be called with
    JH> types instead of classes.  I suppose the name lead me in the
    JH> wrong direction.  I had the silly idea that it only applied to
    JH> instances <0.1 wink>.

    JH> So it comes as little surprise to me that there is a lot of
    JH> code executed in, e.g., the test suite that does comparisons
    JH> on types.

    JH> In the Lib directory, there are 63 files that use == and the
    JH> builtin type function.  (Simple grep.)  A total of 139
    JH> instances of this idiom.  A cursory scan suggests that most of
    JH> the call are things like type(obj) == type('').

Even without the forward-looking insight that types are classes
<wink>, I think type comparisions should have been done with `is' and
not ==.  So old school type comparisons should have been done as

    type(obj) is StringType

whereas new school type comparisons should be done as

    isinstance(obj, StringType)

With Python 2.1 == is naturally, slower than `is', but isinstance()
comes in somewhere in the middle.

563897.802881 is comparisons per second
506827.201066 == comparisons per second
520696.916088 isinstance() comparisons per second

-Barry

-------------------- snip snip --------------------
from types import StringType
import time
r = range(1000000)

def one(r=r):
    x = 'hello'
    t0 = time.time()
    for i in r:
        type(x) is StringType
    t1 = time.time() - t0
    print len(r) / t1, 'is comparisons per second'

def two(r=r):
    x = 'hello'
    t0 = time.time()
    for i in r:
        type(x) == StringType
    t1 = time.time() - t0
    print len(r) / t1, '== comparisons per second'

def three(r=r):
    x = 'hello'
    t0 = time.time()
    for i in r:
        isinstance(x, StringType)
    t1 = time.time() - t0
    print len(r) / t1, 'isinstance() comparisons per second'


one()
two()
three()