Re: [Jython-users] Profiling Jython code

[D-Man <ds...@ri...>]

On Thu, Apr 05, 2001 at 10:57:40AM -0300, German Morales wrote:
| Hi everybody,
...
| I'm looking for a tool that show me the time elapsed in 
| each Jython function or method.
| 
| Having found nothing in a quick search on Google and with
| the same result in this lists, I've made some changes to
| PyTableCode.java, so when I execute 'jython Test.py',
| where Test.py is:
...
|
| I get this result:
| 
| ## ----------------------------------------
| begin - time:986420318682 - code:?
|  begin - time:986420318685 - code:TestClass
|  end - time:986420318686
|  begin - time:986420318686 - code:createFunction
|   begin - time:986420318687 - code:__main__.TestClass.__init__
|   end - time:986420318687
|  end - time:986420318687
|  begin - time:986420318688 - code:__main__.TestClass.printFunction
|   begin - time:986420318691 - code:__main__.TestClass.__str__
|   end - time:986420318697
|  end - time:986420318701
| end - time:986420318701
| ## ----------------------------------------
| 
| The numbers are java.lang.System.currentTimeMillis() calls.
| 
| The next step is to write a program to analize this output.
| 
| But, before going on with my efforts, I wish to ask you some
| questions.
| 
| 
| 0. There is some tool already done to do this job?

I don't know of any.

| 1. If not... What do you think of my aproach?

It is good, but perhaps you could make it easier on yourself.  If it
isn't too difficult with your framework, I would recommend trying
something like the following :


begin = java.lang.System.currentTimeMillis()
some_func_that_may_take_some_time()
end   = java.lang.System.currentTimeMillis()

print "some_func took %d ms" % ( end - begin )


Seeing the difference in time is more interesting (useful) than the
actual start/end times.  I expect you were planning on writing a tool
to parse your existing output and do the subtraction there.  It may be
easier to do that right in the test framework.

| 2. Is someone else interested in profiling jython code?

Sure.  It would be nice to see just how slow (or fast ;-)) it really
is.  It is also very useful when needing to present data to superiors.

A little while back I used Jython to run some performance tests on a
method in our Java runtime that was known to be slow.  I made some
modifications and ran the profiling again.  It was a custom profiling
very similar to the above snippet.  While the numbers weren't exact
(they never are anyways) it gave a good approximation.  Jython also made
checking the accuracy of the modifications easier.  For my final
report I used the same (custom) benchmark application written in Java
that was used for the initial report.

-D