These are chat archives for dropbox/pyston

23rd
May 2016
Dong-hee Na
@corona10
May 23 2016 03:58
@kmod If you ‘re okay Can I handle this issue #1148 and #1099 with docs?
Nathaniel J. Smith
@njsmith
May 23 2016 04:47
is the numpy you're testing still some unknown out-of-date version? might want to double-check before writing headlines :-)
Kevin Modzelewski
@kmod
May 23 2016 04:50
It should be 1.11
we're seeing fewer tests run though than on your CI
I don't think we're doing anything to reduce the number... do you have any ideas about why this might be happening?
yeah, we just clone v1.11.0
Nathaniel J. Smith
@njsmith
May 23 2016 04:53
oh hmm
Nathaniel J. Smith
@njsmith
May 23 2016 05:12
There could be some configuration thing, like maybe the ci is running "slow" tests and you aren't?
It looks like the way the ci runs the tests is to install numpy and then run tools/test-d
tools/test-installed-numpy.py
I'd try running that, and if there's still a mysterious difference I guess I'd run with -v under both CPython and pyston, and then diff the output :-)
Kevin Modzelewski
@kmod
May 23 2016 05:34
hmm, that's the script I'm running
and CPython reports the same number of run tests
ie lower than your setup
I'll guess I'll try running in on your master just to remove that variable
Kevin Modzelewski
@kmod
May 23 2016 06:22
I wonder if it's something like we don't have a fortran compiler installed
Sun
@Daetalus
May 23 2016 06:25
Sorry, I left that message and then go to bed.
Sun
@Daetalus
May 23 2016 11:27
Hi, @kmod I saw you assigned the _ADDDOC stuff to yourself on trello. Would you mind to tell the current situation?
Sun
@Daetalus
May 23 2016 12:31

Hi @njsmith You said check module.__dict__ is not important, so for this test:

builtins = getattr(__builtins__, '__dict__', __builtins__)
assert_(isinstance(builtins, dict))

Simply remove it or submit a patch like this:

import platform
import type
if platform.python_implementation == 'pyston':
    builtins = getattr(__builtins__, '__dict__', __builtins__)
    assert_(isinstance(builtins, types.AttrwrapperType))

Thanks!

Kevin Modzelewski
@kmod
May 23 2016 15:39
@Daetalus re _ADDDOC, it looks like numpy is trying to directly assign to the fields of some objects
specifically, to add a __doc__ string
I think most of our objects are similar to CPython's so we can just enable the structt definitions
I think there's one that's a little bit different and we might want to change
Nathaniel J. Smith
@njsmith
May 23 2016 15:55
@Daetalus: please no pyston check, I just meant rewriting that code so that it doesn't need the explicit type check
I think there are a few ways to do it
But the purpose of the code in question is just to handle both the case where that var is a module and when it is a dict, any modification that works on both is fine. Or maybe it'd be better to remove the use of that variable entirely (it's documented as implementation dependent, after all
)
Sun
@Daetalus
May 23 2016 15:59
Thanks @njsmith and @kmod !
Nathaniel J. Smith
@njsmith
May 23 2016 15:59
@kmod : yeah, this doc string assignment thing is terrible, the pypy folks have been complaining about it too. It really should just be removed from numpy entirely...
@kmod : and ohhhh yeah, you do need a Fortran compiler installed for a bunch of tests
Numpy ships a Fortran wrapper generator, so its whole test suite is probably getting skipped
Marius Wachtler
@undingen
May 23 2016 16:11
mmh looks like github has some problems...
my newest PR keeps up disappearing from the website...
Sun
@Daetalus
May 23 2016 16:12
I encountered this once.
Nathaniel J. Smith
@njsmith
May 23 2016 16:24
It turns out "eventually consistent is sometimes
just another way of saying "inconsistent"
... Also the phone version of gitter's webpage is really annoying, sigh.