Log file Viewer

[07:19:06] <pombreda> d0c_s4vage: actually 3958 was not a real bug after all, but just incorrect markers that setup.py would accept and pip would choke on. I just closed it.

[07:20:15] <ronny> pombreda: can you report a issue to setuptools?

[07:20:30] <pombreda> ronny: good point. will do

[07:20:59] <pombreda> ronny: they should use the same markers parsing/validation code ;)

[07:21:17] <ronny> yep

[07:21:52] <ronny> i hope my new topic on distutils-sig will come in handy for pip/setuptools/others

[07:41:06] <pombreda> ronny: you mean "[Distutils] [proposal] module dependency specification and overriding for packages" https://mail.python.org/pipermail/distutils-sig/2016-September/029659.html

[07:41:09] <pombreda> ?

[07:44:42] <ronny> pombreda: yes, as son as you start to vendor libs with deps, the world breaks

[07:44:46] <ronny> *soon

[07:46:06] <pombreda> ronny: I grok the general issue, but I find it hard to visualize short of an example. Let me reply on the list

[07:51:52] <ronny> pombreda: im nto sure what kind of example you need, a lib with a dependency perhaps?

[07:52:48] <pombreda> any concrete example would do, to highlight a bit more what is a practical problem you face

[07:53:54] <pombreda> ronny: unless you think that this is self evident, in which case this is just me and my crass lack of knowledge

[07:59:38] <ronny> pombreda: i answered with what i think fits, please tell me if it comprehensible now (i would like everyone to understand, even those that didnt yet have to dig deep into those problems)

[08:00:20] <pombreda> checking

[08:03:09] <pombreda> ronny: got it. clear now. yet the Linux distros de-vendoring things and breaking packages is something that I always felt a problematic habit.

[08:03:52] <ronny> pombreda: they are in the right at some level, they probvide hotfixes before upstreams in some cases

[08:03:52] <pombreda> ronny: newer distro like nix and giux have ways to avoid breakages when doing so at least.

[08:04:03] <pombreda> *guix

[08:04:28] <ronny> pombreda: guix/nix have not even the attempt of a reasonable solution

[08:05:01] <ronny> in fact they just dont devendor

[08:05:08] <pombreda> ronny: yep

[08:05:38] <ronny> but the fact that packages need to vendor libraries to begin with points to an complete failure of the languages module system

[08:05:47] <ronny> and in python's case its bascially never fixable

[08:05:49] <pombreda> or rather they multiply the devendoring if needed, hence avoid the one version for everyone problem of other distros

[08:07:05] <pombreda> ronny: complete failure of the languages module system--> yes, that or the need for a packager to provide something that just works! but that is otherwise a python issue and module/packages imports

[08:08:00] <pombreda> ronny: npm/node handles it somehow in quirky way with duplicated node_modules trees and multiple versions

[08:10:47] <ronny> pombreda: bascially note is the other extreme

[08:10:51] <ronny> *node

[08:11:12] <pombreda> I could envision a package installer and import system that would honor "requires" and adjust the code accordingly (with a dynamic namespacing of sorts) such that each packages gets what it needs

[08:11:13] <ronny> pombreda: pythohns module system is not fixuable because we do not have encapsulation

[08:11:37] <ronny> all dependencies leak into the inferdace of a module

[08:11:42] <pombreda> ronny: it could be fixed by a smarter installer

[08:11:47] <ronny> thus different versions of dependencies leak around

[08:11:47] <pombreda> ronny: true

[08:11:56] <ronny> it is a problem that is fundamentally not fixable

[08:12:09] <pombreda> :|

[08:12:56] <ronny> pombreda: it is simply impossible to have miltiple versions of a library in a process and act correct with a type system like python's

[08:14:18] <pombreda> ronny: agreed, unfortunately

[16:07:04] <d0ugal> so, something is going on with setuptools?

[16:48:24] <sigmavirus> d0ugal: ?

[20:41:07] <d0ugal> sigmavirus: Nevermind :)

[20:41:27] <d0ugal> sigmavirus: A release was deleted, that's all