Comments on: Optimisation – it’s sometimes needed. http://www.teamrubber.com/blog/optimisation-its-sometimes-needed/ Team Rubber talks on the Internet in a blog Sat, 31 Jul 2010 03:38:02 +0000 http://wordpress.org/?v=2.8.5 hourly 1 By: Tim Wintle http://www.teamrubber.com/blog/optimisation-its-sometimes-needed/comment-page-1/#comment-9948 Tim Wintle Tue, 10 Nov 2009 18:11:21 +0000 http://www.teamrubber.com/blog/?p=1004#comment-9948 ... because I'm leaving subclasses the option of setting _batchlist to a different (list-like) class in case I want to do lazy joins in the future. (there is actually a descriptor called "batchlist" that lets you get/set the _batchlist variable that I've missed out from this) For the same reason, I only cast batchlist to (PyObject *) -"object" in Cython. I suspect it would be faster by importing listobject.h and casting direct to the list type (Cython's generated C is actually quite messy around calling the "append" method in case the method doesn't exist) - but Cython makes use of a lot of gcc optimisations so the effect shouldn't be too great. … because I’m leaving subclasses the option of setting _batchlist to a different (list-like) class in case I want to do lazy joins in the future.

(there is actually a descriptor called “batchlist” that lets you get/set the _batchlist variable that I’ve missed out from this)

For the same reason, I only cast batchlist to (PyObject *) -”object” in Cython.

I suspect it would be faster by importing listobject.h and casting direct to the list type (Cython’s generated C is actually quite messy around calling the “append” method in case the method doesn’t exist) – but Cython makes use of a lot of gcc optimisations so the effect shouldn’t be too great.

]]> By: Richard Barrell http://www.teamrubber.com/blog/optimisation-its-sometimes-needed/comment-page-1/#comment-9935 Richard Barrell Tue, 10 Nov 2009 17:54:53 +0000 http://www.teamrubber.com/blog/?p=1004#comment-9935 The improvement there is quite impressive, especially when it's almost as great even when accessed through the subclass machinery. Next question: since self._batchmax is known in advance, why not use an array of PyObject pointers instead of a Python list for _batchlist? That way, there'd be no array resizing as extra elements get added. The improvement there is quite impressive, especially when it’s almost as great even when accessed through the subclass machinery.

Next question: since self._batchmax is known in advance, why not use an array of PyObject pointers instead of a Python list for _batchlist? That way, there’d be no array resizing as extra elements get added.

]]>