PMXBOT Log file Viewer

Help | Karma | Search:

#pypa logs for Tuesday the 12th of February, 2019

(Back to #pypa overview) (Back to channel listing) (Animate logs)
[03:10:23] <njs> hrw: so yeah the blocker to uploading aarch64 wheels to pypi is that we don't have an ABI spec. Specifically, we need some way for pip to figure out "can I expect this aarch64-tagged wheel to work on this system?", and for developers to figure out "how do I build this so ti works on all the systems where pip will isntall it"
[03:10:58] <njs> everyone would be happy to support it, but answering those questions requires some specific technical expertise about the platform and its ABIs, which we don't have
[15:00:17] <dude-x> in setup.py is there a way to specify sqlalchemy 1.3 which would normally require the --pre flag in setup_requires?
[17:27:26] <tos9> dude-x: if you specify a prerelease version, it will use it
[17:27:49] <dude-x> i had to set it to sqlachlemy == 1.3.0b3
[17:56:34] <dude-x> hmmm my virtualenv i just created is missing stuff
[18:24:40] <dude-x> virtualenv is not copying all the files properly
[18:28:02] <dude-x> into lib/python/
[18:30:33] <dude-x> nevermind its working by design.
[18:41:27] <snatcher> is there a way to upgrade all packages with pip?
[20:19:56] <ngoldbaum> today on twitter: dstufft: "i have a cool idea for thing", a bunch of people with financial stakes in python packaging being annoying: lots of FUD
[20:30:44] <tos9> what drama did I miss and do I care (probably not, but dstufft you are great)
[20:30:47] <dude-x> snatcher other than scripting, no
[20:31:32] <dstufft> tos9: it's mostly some conda adjacet people getting really confused at me for saying I started working on a rough draft for a new tool that might solve. alot of pain points
[20:31:47] <dstufft> and then deciding that meant I wastrying to eat into conda's marketshare or something?
[20:32:10] <dstufft> idk, they keep trying to talk about pip vs conda, but my thing is more project management
[20:32:22] <dude-x> i thought conda is more about making it easy to install numpy related libs easily on windows (and mac)
[20:32:32] <tos9> dstufft: I see.
[20:32:36] <snatcher> dude-x: what's the reason?
[20:33:34] <dude-x> snatcher well, you can have a requirements file that has no versions pinned and just pip install -U -r req
[20:34:01] <tos9> dstufft: As a non-conda user who finds lots wrong with package management I probably would love new rough drafts of tools that solve problems!
[20:34:08] <tos9> dstufft: So thanks for trying!
[20:35:32] <snatcher> *what's the reason to not include upgrade functionality into package manager?
[20:37:17] <dude-x> snatcher pip install -U package works
[20:37:55] <dude-x> you can use pipenv that makes this easier too
[20:40:42] <snatcher> dude-x: i know i can upgrade specific package, but what's the problem to update all packages automatically (thing that 99% of package managers do), is there any reason to not add that functional, don't think it's so hard to implement
[20:41:31] <dude-x> i am not a pip maintainer. but as a professional, you have to be conservative in updating all your packages because things break due to bugs, or api changes.
[20:42:18] <dude-x> i personally update all my packages locally, but for the stuff that runs on CI, it's all pinned until I need to upgade stuff
[20:42:38] <dude-x> and then, i have to spend some development time to test the upgrades and review the changes
[21:03:32] <ngoldbaum> dstufft: yes, your idea looks really cool, and is coming at the same time greg szork is doing cool stuff with rust and mercurial
[23:13:10] <njs> snatcher: the only reason there's no upgrade-all-packages command in pip is because of a technical logjam – pip's current tools for checking whether packages are compatible with each other are limited/buggy in certain ways that an upgrade-all command would reveal
[23:13:37] <njs> snatcher: so the "add an upgrade-all command" todo item has gotten stuck behind the "rewrite pip's compatibility-checking code" todo