Author archives: Mike

There is a thread on Python-list about removing the "cmp" parameter to Python's sort. One little tidbit I had missed was the introduction of a stable sort guarantee with Python 2.2. I've got lots of older code that uses cmp, so I figured I'll just start rewriting those as I come across them... later that ...

As I've been listening to PyCon lectures in the background as I work, I'm struck by the number of times people discussing PyOpenGL refer to the OpenGL.array.vbo module positively.  It's actually one of those pieces of the API where I have broken one of my rules for the project; "Don't extend the API".

Over the ...

Add a SquareMap-style visualization to accounting software, instant visualization of your entire company's expenses, revenue, budget, etceteras.  Perfect for including in reports to CEOs, Shareholders and the like.  For bonus marks, make it explorable, with a web-based tool to which I upload simple tree-structured data-sets.

Unscheduled day off work today... which means... RunSnakeRun (and SquareMap) got quite a few performance improvements when handling Meliae dumps.  It still takes 3-4 minutes to load a 16MB capture (that is, an dump that captured 16MB of objects) but once loaded it is reasonably close to interactive.  I haven't got time to work on ...

I've just released the latest version of RunSnakeRun.  Only real change is to better handle meliae dumps in the sense of giving more reasonable data.  The views are now reasonably useful, but they are extremely slow for any real-world application.

I've released (on PyPI) versions of RunSnakeRun and SquareMap that support Meliae dumps... but I'm not really satisfied with the functionality.  Basically the dumps seem to have 2-3MB of un-accounted-for objects in them.  In my trivial test that's about 10x more memory than can be reached from module references.  This small dump of ~13,000 objects ...

Got Meliae exploration working in RunSnakeRun this evening.  A few enhancements to the SquareMap control as well.  The data-sets are a bit too large on real-world memory-captures so performance becomes a serious issue... will have to do some profiling :) .  Oh well, something for another day.

Dave Malcolm's talk at PyCon 2011 mentioned Meliae, which is a memory-crawler that stores out a description of all of the objects in RAM when triggered.  It's Python-level, so not something you can use with Dave's gdb stuff, but it still sounds pretty useful.

Anway, while the summary stuff it produces is okay, but ...

Want to see the output of your nosetests run in a format that you can hand to someone non-technical (who doesn't mind it being ugly)?  Here's a little script to convert the result of nosetests --with-xunit to an ugly, but functional HTML file.  I can't say I really see this as a long-term solution, as ...

Another day playing with PyPy. First up was a pleasant surprise in that the 2x slowdown reduced with the currently nightly build[1], bringing performance up from 40,000cps to ~60,000cps.

After that, applied a micro-optimization that got me about a 6% speedup; I eliminated the use of a state "struct object" (object that just used regular ...