pmxbot IRC Log Viewer

[06:36:53] <pradyunsg> dude-x: what’s the error message?

[10:55:48] <hrw> hello

[11:04:44] <hrw> filled https://github.com/pypa/warehouse/issues/7980

[11:05:08] <hrw> I hope that answer will be other then WONTFIX or KTHXBYE.

[11:55:51] <ngoldbaum> i think that means py-spy would no longer work

[11:57:10] <hrw> ngoldbaum: ?

[11:58:08] <ngoldbaum> py-spy is written in rust so it lives on crates.io, still it has a python binary distributed as a wheel on pypi

[11:59:00] <ngoldbaum> https://pypi.org/project/py-spy/#files

[11:59:34] <ngoldbaum> also how would pypi verify the wheels come from the uploaded tarball?

[11:59:57] <ngoldbaum> what’s stopping someone from uploading an empty tarball or something

[13:13:25] <dude-x> pradyunsg i don't remember exactly but it said for many packages: unable to install package 'blah' because module 'wheel' was not found.

[13:14:13] <pradyunsg> Ah, try now with pip 20.1.1 -- when that message is printed and the contents of that message were improved.

[13:14:17] <pradyunsg> dude-x: ^

[13:14:40] <hrw> ngoldbaum: you can still push wheel. just push source

[13:14:56] <dude-x> pradyunsg okay i'll try pip installing requirements again

[13:14:58] <ngoldbaum> but there is no source

[13:15:02] <ngoldbaum> that’s the thing

[13:15:25] <hrw> ngoldbaum: do you publish aarch64 wheel? ppc64le wheel? riscv wheel/

[13:15:26] <hrw> ?

[13:15:53] <ngoldbaum> what do i have to do with anything?

[13:16:08] <ngoldbaum> the py-spy devs can publish whatever they want to publish

[13:16:10] <hrw> I run aarch64 system and want to use your py-spy.

[13:16:26] <hrw> but no aarch64 wheel, no source so can not

[13:16:28] <ngoldbaum> ok, start with the rust sources

[13:16:36] <hrw> and can not use all projects which use it

[13:16:49] <ngoldbaum> i just

[13:17:00] <hrw> ngoldbaum: make it work with 'pip install py-spy'? fail if no rustc etc

[13:17:23] <ngoldbaum> just trying to point out that there are real-world usecases that your requirement would break

[13:17:39] <ngoldbaum> go ahead and wait for someone from pypa to respond to your issue

[13:17:44] <hrw> I may consider them invalid ;)

[13:17:59] <hrw> but understand them too

[13:18:46] <hrw> I had too many wasted days because someone pushed wheel only to pypi and broke all my CI jobs

[13:19:11] <mgedmin> something something technical solutions for social problems

[13:19:13] <hrw> so instead of doing my work I had to check why they failed, find which package, where to report bug etc

[13:19:30] <ngoldbaum> seems like a bug in protobuf’s release process to me

[13:19:41] <ngoldbaum> not really a pypi problem per se

[13:20:11] <ngoldbaum> especially if the “solution” breaks a lot of other things that use the ability to only publish binaries for a project

[13:20:53] <ngoldbaum> iirc tensorflow does something similar with their wheels

[13:21:01] <ngoldbaum> all these google projects :p

[13:22:43] <McSinyx[m]> we can, on the other hand, enforce projects on PyPI to be free/libre software though

[13:23:04] <McSinyx[m]> AFAIK we don't have any rules about the licensing of packages uploaded to PyPI

[13:23:46] <McSinyx[m]> if we ensure that they're libre software, the license will basically cover the reproducible source part

[13:23:54] <ngoldbaum> i mean these are all free/libre projects

[13:24:02] <ngoldbaum> the source is just available elsewhere

[13:24:05] <ngoldbaum> not on pypi

[13:24:17] <ngoldbaum> only using pypi for the binary distribution via pip

[13:24:37] <McSinyx[m]> ah got it, so it's rather just inconvenient to automate?

[13:24:56] <McSinyx[m]> *automate building from source

[13:24:57] <ngoldbaum> for protobuf it seems like someone forgot to upload the sdist this time

[13:25:01] <hrw> the prbolem starts when binaries do not cover platforms

[13:25:17] <ngoldbaum> for the other’s it’s a conscious decision

[13:26:17] <dude-x> pradyunsg i reran the install and saw no errors. so i guess force installing wheels works

[13:26:47] <McSinyx[m]> in alternative to enfocing sdist (which is not necessarily buildable using e.g. pip), we can force them to link to source

[13:26:53] <McSinyx[m]> but even that is up to debate

[13:29:48] <McSinyx[m]> ngoldbaum: just looked up protobuf, i also think they did not considered uploading sdist

[13:30:30] <hrw> McSinyx[m]: and did for each previous ones

[13:30:30] <McSinyx[m]> dude-x: do you still need to force-wheel?

[13:30:45] <hrw> McSinyx[m]: current version is not installable

[13:30:47] <ngoldbaum> yeah, like i said, a bug in their release process, or they’ve arbitrarily decided to stop now on a bugfix release

[13:30:56] <dude-x> McSinyx[m] no it was just a one time thnig

[13:30:57] <ngoldbaum> which would be strange

[13:33:01] <McSinyx[m]> dude-x: I'm asking because I'm curious, don't really know why that happens, I'm sorry

[13:34:57] <dude-x> McSinyx[m] ah. i activated an old venv in pyenv, wanted to make sure i updated my dependencies that may have changed in the repo. ran pip install -r requirements and got a ton of wheel not found errors.

[13:50:50] <dude-x> ooh latest pip removes multiprocessing. it was pretty slow on my mac for non python 3.8 checks for outdated.

[13:52:28] <McSinyx[m]> dude-x, see https://github.com/pypa/pip/issues/8169 and linked issues

[13:53:33] <dude-x> McSinyx[m] do you need more than one core for multithreading to work properly for IO bound wokr?

[13:55:09] <McSinyx[m]> I don't think so, but we're uncertain about edge cases like on android (no sem_open)

[13:55:23] <ngoldbaum> no, especially for things waiting on the network

[13:55:31] <dude-x> fair enough.

[13:55:35] <dude-x> what does sem_open do?

[13:55:38] <hrw> hm. multithreaded pip...

[13:55:44] <dude-x> open a semaphore?

[13:55:51] <McSinyx[m]> yes

[13:56:05] <McSinyx[m]> multiprocessing uses it

[13:56:06] <hrw> On-line CPU(s) list: 0-45

[13:56:14] <dude-x> huh android doesn't provide semaphors? that's interesting.

[13:56:25] <McSinyx[m]> (the thread pool used in pip list is from multiprocessing dummy)

[13:56:49] <McSinyx[m]> dude-x: I'm not sure, I only know that sem_open is not supported

[13:57:04] <McSinyx[m]> it's generally a new-ish thing (10y or so)

[13:57:46] <McSinyx[m]> https://bugs.python.org/issue3770

[17:27:56] <dstufft> McSinyx[m]: PyPI does not require that software uploaded to it be under some F/OSS license

[17:29:27] <dstufft> We do have pretty specific legal rquirements that must be satisified when you upload something to PyPI

[17:29:58] <dstufft> https://pypi.org/policy/terms-of-use/

Log file Viewer

Help | Karma | Search:

#pypa logs for Tuesday the 19th of May, 2020