You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@thrift.apache.org by "Will Pierce (JIRA)" <ji...@apache.org> on 2011/03/23 05:07:05 UTC

[jira] [Commented] (THRIFT-1107) improvement for compiler-generated python for 'None' object comparisons

    [ https://issues.apache.org/jira/browse/THRIFT-1107?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13009988#comment-13009988 ] 

Will Pierce commented on THRIFT-1107:
-------------------------------------

Output from benchmark script (test_isnotnone.py attached):

% for i in python2.4 python2.7 python3.1; do $i ./test_isnotnone.py 125; done 
h2. Testing python v2.4.6, 125.0 million iterations
|| Code || Time Per Iteration (microseconds) || Total Test Elapsed Time (seconds)
|       x != None | 0.08892 usec/iteration |  11.115 total sec elapsed
|   x is not None | 0.05089 usec/iteration |   6.361 total sec elapsed

h2. Testing python v2.7.0, 125.0 million iterations
|| Code || Time Per Iteration (microseconds) || Total Test Elapsed Time (seconds)
|       x != None | 0.06088 usec/iteration |   7.611 total sec elapsed
|   x is not None | 0.03261 usec/iteration |   4.077 total sec elapsed

h2. Testing python v3.1.2, 125.0 million iterations
|| Code || Time Per Iteration (microseconds) || Total Test Elapsed Time (seconds)
|       x != None | 0.08650 usec/iteration |  10.813 total sec elapsed
|   x is not None | 0.02976 usec/iteration |   3.720 total sec elapsed



> improvement for compiler-generated python for 'None' object comparisons
> -----------------------------------------------------------------------
>
>                 Key: THRIFT-1107
>                 URL: https://issues.apache.org/jira/browse/THRIFT-1107
>             Project: Thrift
>          Issue Type: Improvement
>          Components: Python - Compiler
>            Reporter: Will Pierce
>            Assignee: Will Pierce
>         Attachments: test_isnotnone.py
>
>
> The python code generator produces several python statements, especially the {{write()}} methods' per-field code, that compares something to None using '{{!= None}}', when it is more efficient to use the '{{is not None}}' expression.
> From what I understand, in python it's almost always true that ({{x \!= None}}) == ({{x is not None}}), but the actual implementation and intent is very different.  The '{{\!= None}}' comparison does a by-value comparison that does much more work than an object identity '{{is not None}}' comparison does.
> The actual performance impact isn't much, but I benchmarked the performance of '{{x is not None}}' to '{{x \!= None}}' and got some interesting results.  In python 2.4, 2.7 and 3.1, it's about 2-3 times as fast to use '{{is not None}}' over '{{\!= None}}'.
> I'll attach a patch to switch to 'is not None', and attach a simple benchmark test script exercising '{{is not None}}' vs. '{{\!= None}}' and post the performance measurements to this ticket.
> These URLs are somewhat relevant about this specific issue in general:
> * http://www.python.org/dev/peps/pep-0008/ (Search for 'singletons', or scroll to 'Programming Recommendations' item 2)
> * http://stackoverflow.com/questions/100732/why-is-if-not-someobj-better-than-if-someobj-none-in-python - Stack Overflow question about the same, though it veers into the cost of typecasting to bool, which isn't relevant here
> * http://jaredgrubb.blogspot.com/2009/04/python-is-none-vs-none.html - a similar experience and test results that match

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira