On Wed, Oct 01, 2008 at 01:55:21AM -0700, Dan Price wrote:

> You get the action item to fix actionbench and fmribench to do whatever
> it is needs to be done with PYTHONPATH :)  Perhaps just printing its
> value would be enough...

Not for me.  When I'm blond, I'm blond.  The difference between my two
PYTHONPATHS:

   
/export/home/dduvall/packaging/pgate/proto/root_i386/usr/lib/python2.4/vendor-packages
   
/export/home/dduvall/packaging/pkg-clone/proto/root_i386/usr/lib/python2.4/vendor-packages

which blend together pretty well.  I need a better device to help me use
the right one.  For the functionality testing, having PYTHONPATH set
automatically has saved my ass many times; perhaps I should investigate
doing the same for the performance testing.

> > As for profiling, it looks like
> > 
> >               libpython2.4.so.1.0`PyObject_Malloc+0xb3
> >               libpython2.4.so.1.0`PyString_FromStringAndSize+0xf4
> >               _actions.so`_fromstr+0x56f
> 
> So, what we expected.  Hmm... well, that points to the idea of
> storing some stuff in C structs, or something.

That could help us with memory footprint, potentially, by not creating
Python strings until we need 'em, but I don't see how we can avoid paying
the time cost of creating those strings, unless we never access them, in
which case we probably shouldn't ever parse the action to begin with.

I think there may be a compromise in delaying creation of all but the key
attribute, but I don't know how much that would actually save us.  Things
to toss around, at least.

Danek
_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss

Reply via email to