On Thu, 2011-12-22 at 10:56 +0100, Antoine Pitrou wrote: > On Thu, 22 Dec 2011 09:44:32 +0000 > Tim Wintle <timwin...@gmail.com> wrote: > > > > 2.5 apps are the speed-critical ones. Our tests showed the performance > > was different enough between 2.5 and 2.6 for me to not update. > > Really? Where's the regression?
I'm not certain - IIRC there were several nice optimisations in 2.6, and I wasn't expecting that when I first looked. I was running code designed for 2.5 under 2.6, so it's likely that with sufficient tweaking for 2.6 I might not have the same result. I tested this specific code with the python builds we have in production, not general python code - I don't mean this as a recommendation that anyone else assume 2.5 is faster for them. I suspect that Stefan's comments about newly added features without the optimisation in python3 might be partially true, but having the extra code to support them (while not using them) might also be part of the cause - ceval.c had over 1K line changes between r25 and r26, including cases for new opcodes, and new opcode predictions etc - it's possible that my code just happens to not follow the most optimal paths. I'm talking about a slow-down of under 10%, but enough that I couldn't justify moving these apps to 2.6 at the time for economic reasons, and pypy would be the main incentive to move this to 2.7. Tim _______________________________________________ Python-Dev mailing list Python-Dev@python.org http://mail.python.org/mailman/listinfo/python-dev Unsubscribe: http://mail.python.org/mailman/options/python-dev/archive%40mail-archive.com
