Re: Performance of PyUnicode Issue #2100

I've implemented the indexing idea now, and it makes most unicode 
operations on long-ish strings a lot faster, without significantly 
harming BMP performance. Two change sets are worth a look. (I'm pushing 
to Bitbucket to avoid a whole series of merges.)

The first adds the idea of index translation, but accomplishes it using 
String methods that have to count code points:
https://bitbucket.org/tournesol/jython-ja/commits/191a9854396d5ba742d4b2d3179647eb632aa6d0#Lsrc/org/python/core/PyUnicode.javaT132

That one is correct but slow. Then the second one provides the index 
that makes access fast.
https://bitbucket.org/tournesol/jython-ja/commits/b96c8402f7baff86b86a002194a3b50d6903a04c

The bug I noticed (in find) is fixed by the sequence of changes, and 
others I've found (replace, endswith, startswith) where arguments 
defining a slice were not dealt with correctly. I think I've found them 
all, but maybe not. A couple of other things I want to do:

1. speed up slicing, using the same technique, where easy. I'll try to 
resist more complex cases, like how we do strided slices, although I 
think there's scope.

2. positively reject strings that include lone surrogates of any kind. 
At present, I only catch leading surrogates without a trailing surrogate.

Shout if any of this looks wrong.

Jeff

Jeff Allen

On 02/09/2014 19:11, Jeff Allen wrote:
> JMH is worth the learning curve. I have quite informative benchmarks 
> now for the table building and index translation. thanks for the pointer.
> Jeff Allen
> On 19/08/2014 08:15, Jeff Allen wrote:
>> Interesting links, thanks. Caliper seems to have a good approach and 
>> explains itself well. I had experimented with the flags it uses to 
>> detect how many repeats might be needed to make recompilation stop. 
>> JMH looks harder to apply. (I found a tutorial elsewhere.) Overall I 
>> quite like that my test is in Python, but I can see uses for these 
>> either in the next phase and where I might have used them in the past.
>> Jeff Allen
>> On 17/08/2014 09:55, Indra Talip wrote:
>>> Jeff,
>>> Two tools that I've seen that have a reasonable reputation for being 
>>> able to create reliable micro benchmarks on the JVM are JMH [1] and 
>>> Caliper [2]. I've not used either but my understanding is that they 
>>> are suited to that sort of benchmarking and handle the vagaries of 
>>> the JIT compiler.
>>>
>>> [1] http://openjdk.java.net/projects/code-tools/jmh/
>>> [2] https://code.google.com/p/caliper/
>>>
>
>
>
> ------------------------------------------------------------------------------
> Slashdot TV.
> Video for Nerds.  Stuff that matters.
> http://tv.slashdot.org/
>
>
> _______________________________________________
> Jython-dev mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/jython-dev

------------------------------------------------------------------------------
Want excitement?
Manually upgrade your production database.
When you want reliability, choose Perforce
Perforce version control. Predictably reliable.
http://pubads.g.doubleclick.net/gampad/clk?id=157508191&iu=/4140/ostg.clktrk

_______________________________________________
Jython-dev mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/jython-dev