These are chat archives for dropbox/pyston

17th
Feb 2016
lesshaste
@lesshaste
Feb 17 2016 05:16
@kmod A post setting out what is new/better/exciting in the pyston work since november would be great :) Also, what still needs to be done and what the plan is for the next few months. My own interest is in performance and the use of python modules such as scipy and scikit-learn. This is a major weakness of pypy for example.
lesshaste
@lesshaste
Feb 17 2016 09:57
@kmod the table 1 from https://fedcsis.org/proceedings/2015/pliks/139.pdf is seriously out of date now with respect to pypy. I just tried pypy and it is faster than cpython in all those tests
Marius Wachtler
@undingen
Feb 17 2016 10:57
:-), pyston got also much better compared to the version they have run when creating the paper
lesshaste
@lesshaste
Feb 17 2016 11:25
@undingen did it get better before 2015 though :) ?
they don't actually state the version numbers which is odd
at least as far I could see
@undingen what is the situation with pyston and numpy?
Marius Wachtler
@undingen
Feb 17 2016 11:33
I think at the time they tested pyston (and with 0.2 at least they mentioned the took some of the benchmarks of this release (strange that the don't mention the pypy version)) was really early so I'm not surprised that it failed to run a lot of tests and performed bad.
I'm not sure about numpy. @Daetalus should be able to give you a better status update :-)
lesshaste
@lesshaste
Feb 17 2016 13:10
@undingen thanks
An Long
@aisk
Feb 17 2016 14:45
Hi @undingen can you help me restart this travis job:https://travis-ci.org/dropbox/pyston/jobs/109877700 , seems gnu’s git server is down before, now it’s recovered.
Marius Wachtler
@undingen
Feb 17 2016 14:55
seams to be still down :-(
An Long
@aisk
Feb 17 2016 14:56
😰
Marius Wachtler
@undingen
Feb 17 2016 15:09
it's up :-)
An Long
@aisk
Feb 17 2016 15:09
cool, thanks!
Thomas Mangin
@thomas-mangin
Feb 17 2016 15:10
travis seems to have issues (they have migrated their infra). I perform some fonctional testing (python does not need compilation) and I have placed timeout to make sure the testing fails if something hangs. The build are all failing atm as they are too slow.
An Long
@aisk
Feb 17 2016 15:12
Seems more people moved to Circle CI?
Marius Wachtler
@undingen
Feb 17 2016 15:21
last week there was also some issue which made the builds fail all the time
Thomas Mangin
@thomas-mangin
Feb 17 2016 15:27
Also using both drone.io and codeship (as free for open projects)
Kevin Modzelewski
@kmod
Feb 17 2016 21:07
ooh Circle CI offers SSH support
which would be nice if/when we have those CI-only bugs show up again
Kevin Modzelewski
@kmod
Feb 17 2016 21:47
Interesting, searching the sqlalchemy docs is an interesting history of incompatibilities they had to fix
here's an interesting one:

[orm] [bug] [pypy] Fixed regression from 0.9.10 prior to release due to #3349 where the check for query state on Query.update() or Query.delete() compared the empty tuple to itself using is, which fails on Pypy to produce True in this case; this would erronously emit a warning in 0.9 and raise an exception in 1.0.¶

References: #3405

and there are a lot of "we didn't close the cursor in these cases and that was an issue for pypy" bugs
Marius Wachtler
@undingen
Feb 17 2016 22:58
oh thats interesting and may help
Thomas Mangin
@thomas-mangin
Feb 17 2016 23:17
I also had the opposite code working under pypy but failing under Python
Kevin Modzelewski
@kmod
Feb 17 2016 23:17
Thomas Mangin
@thomas-mangin
Feb 17 2016 23:17
It was related to some use of new if I recall correctly
Also Python has some bugs it can not fix due to backward compatibility in some form of object caching - I could dig the details if relevant
Kevin Modzelewski
@kmod
Feb 17 2016 23:19
we're definitely interested in any issues that we're having :)
Thomas Mangin
@thomas-mangin
Feb 17 2016 23:19
I will try to find the time and report here