These are chat archives for dropbox/pyston

18th
Jan 2016
Marius Wachtler
@undingen
Jan 18 2016 09:52
I took a brief look at some numpy code and it looks like numpy array type also depends on refcounting :-(.
Without immediately calling into the array_dealloc we will return the wrong values...
Marius Wachtler
@undingen
Jan 18 2016 09:59
if I manually trigger a gc collection the values will be correct afterwards :-D
Rudi Chen
@rudi-c
Jan 18 2016 18:09
Ok. Since it's merge I'll go update the other NumPy PR then.
Sun
@Daetalus
Jan 18 2016 18:11
:thumbsup:
Nathaniel J. Smith
@njsmith
Jan 18 2016 19:14
oh ick, you're absolutely right about that numpy thing. ugh.
Nathaniel J. Smith
@njsmith
Jan 18 2016 19:25
@undingen : probably there is no solution except to modify numpy to get rid of UPDATEIFCOPY. there's no intrinsic reason it has to be done the way it is, and in fact it'd probably make some of the code clearer and less magical, but finding and cleaning all that stuff up may take some work.
Marius Wachtler
@undingen
Jan 18 2016 19:28
:-( but thanks for your expert opinion. We are still thinking about switching to ref counting so this issue may get resolved. I'm sadly not having time to change the numpy implementation at the moment.
Nathaniel J. Smith
@njsmith
Jan 18 2016 19:35
yeah, I hear that :-/
I will poke the pypy folks also, they are also currently trying to get their C-API to run numpy and will surely run into the same problem very soon, maybe they will be inspired to do something about it :-)
Marius Wachtler
@undingen
Jan 18 2016 19:42
:-)
Nathaniel J. Smith
@njsmith
Jan 18 2016 19:53
well, filed an issue to at least track the problem: numpy/numpy#7054
Marius Wachtler
@undingen
Jan 18 2016 19:55
Thanks for filling the issue!
datnamer
@datnamer
Jan 18 2016 21:43
Hey everyone. A quick and probably common question: Any chance pyston will get ported to python 3 in the medium future?