PyPy dev here. Could you give us some info on how to help you more, for instance, a small benchmark of where we are still slow? We love to hear how PyPy is being used "in the real world"
I was too optimistic in my eyeballed assessment, but a real measurement shows there's still a good improvement. I gave my program a typical workload and tested with both PyPy 2.4.0 and 2.5.0, and the result was 13 minutes 31 seconds on 2.4.0, and 11:22 on 2.5.0. That's great!
What would be the best way to profile this thing under PyPy to see where it's spending time? I'm barely familiar with Python profiling in general, and totally unfamiliar with PyPy specifically.
As for how I'm using it, I belong to a glider club. For every weekend day we operate (which is every weekend from around the end of February through mid-December) we need to assign four club members to carry out various tasks for the day. (Specifically, we need a designated tow pilot, an instructor, a duty officer, and an assistant.) I'm the scheduling guy, and I wrote a Python program to generate the schedules for me. I put in various constraints (when people are unavailable, what days they prefer, etc.) and then the program uses a hill-climbing algorithm along with a bunch of scoring routines to find a good schedule. The actual workload operates on Schedule objects, which just contain a list of dates and the members assigned to each position on each date. Then I make a ton of duplicates, each with one adjustment, score them all, pick out the best, repeat. I can also optionally go for two adjustments, which takes much longer but gives better results, and that's what ends up taking 10+ minutes as above.
Hey, if your code is Open Source, then I would be willing to look at your code and help you optimize it. Hit me on #pypy on IRC or just send me a mail at fijall at gmail.
