HN Explorer

9 comments

[ 3.8 ms ] story [ 34.6 ms ] thread
rcarmo
Hmmm. Time to try some of my Cython stuff again, then :)

Update: the peewee ORM's playhouse extensions still break :(

staticautomatic
I wonder what the benchmarks would look like for Pandas on pypy compared to regular python.
dr_zoidberg
> We changed timeit to now report average +/- standard deviation, which is better than the misleading minimum value reported in CPython.

While I don't agree with many of the CPython devs ideas/decisions/crawling-pace-decision-making, I don't like the tone with which PyPy devs refer to them most of the time. It's like a rivalry which is ignored from the CPython side, even unacknowledged. And PyPy is not the Holy Grail of Python, it's just another implementation which runs into its own problems on some places[0], so it feels weird seeing them with kind of constant badgering of "the other guys".

The feature looks interesting, still if you don't like a particular feature/profiler/timer for Python, there's nothing stoping you from making your own[1], like kernprof which is amazing.

[0] for example, if you don't follow RPythons "way of doing things" you can end up with even slower performance than CPython -- I know because I've done it!

[1] https://github.com/rkern/line_profiler#id7

dripton
I don't see anything particularly rude about this statement. I think you're overreacting to something that isn't there.
stuaxo
If something is slower in CPython than in Pypy, I'm pretty sure they are usually up for treating this as a bug.

Things that are slow because of warmup used to not be the focus, but there is some focus there now too.

lomereiter
Also, minimum value is in some ways actually better, as argued by Alexandrescu in one of his talks [0], as it effectively removes various sources of noise from the measurement. Not sure to what degree this argument is applicable to a language with a GC, though.

[0] https://www.youtube.com/watch?v=vrfYLlR8X8k (starting from 15 min)