4 ms·

PyPy2.7 v5.6 released – stdlib 2.7.12 support, C-API improvements, and more

61 points by mattip 10y ago

scarygliders 10y ago
Is there a reason PySide fails other than the lack of cmake, as the error alludes to here : http://packages.pypy.org/##PySide http://packages.pypy.org/##PySide ?
rcarmo 10y ago
Hmmm. Time to try some of my Cython stuff again, then :) Update: the peewee ORM's playhouse extensions still break :(
- stuaxo 10y ago
  They are generally pretty friendly, it may be worth bringing it up on the mailinglist, the advice may be to open a bug up then.
  - rcarmo 10y ago
    It's apparently been fixed by not building the C versions under pypy: https://github.com/coleifer/peewee/issues/1118 https://github.com/coleifer/peewee/issues/1118
staticautomatic 10y ago
I wonder what the benchmarks would look like for Pandas on pypy compared to regular python.
dr_zoidberg 10y ago
> 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 https://github.com/rkern/line_profiler#id7
- dripton 10y ago
  I don't see anything particularly rude about this statement. I think you're overreacting to something that isn't there.
- stuaxo 10y ago
  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 10y ago
  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 https://www.youtube.com/watch?v=vrfYLlR8X8k (starting from 15 min)

61 points by mattip 10y ago

Is there a reason PySide fails other than the lack of cmake, as the error alludes to here : http://packages.pypy.org/##PySide http://packages.pypy.org/##PySide ?

rcarmo 10y ago

Hmmm. Time to try some of my Cython stuff again, then :) Update: the peewee ORM's playhouse extensions still break :(

stuaxo 10y ago
They are generally pretty friendly, it may be worth bringing it up on the mailinglist, the advice may be to open a bug up then.
- rcarmo 10y ago
  It's apparently been fixed by not building the C versions under pypy: https://github.com/coleifer/peewee/issues/1118 https://github.com/coleifer/peewee/issues/1118

staticautomatic 10y ago

I wonder what the benchmarks would look like for Pandas on pypy compared to regular python.

dr_zoidberg 10y ago

> 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 https://github.com/rkern/line_profiler#id7

dripton 10y ago
I don't see anything particularly rude about this statement. I think you're overreacting to something that isn't there.
stuaxo 10y ago
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 10y ago
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 https://www.youtube.com/watch?v=vrfYLlR8X8k (starting from 15 min)