pmxbot IRC Log Viewer

[00:03:06] <dstufft> _habnabit: yea, "0.1" matches all local versions of 0.1

[00:03:28] <dstufft> Not sure it's possible to exclude 0.1 but allow local versions of 0.1 without pinning to that local version

[00:03:31] <_habnabit> dstufft, so there's no way to match/filter on the upstream version?

[00:03:47] <dstufft> I mean you can do ==0.1.0+upstream.2

[00:03:54] <dstufft> but that's it I think

[00:04:08] <dstufft> agronholm: it's reindexing, hopefully that fixes it well enough for tonight

[00:04:16] <dstufft> I'm not feeling great

[00:04:32] <agronholm> I have a conditional dependency in extras_require to install typing only for python 3.4

[00:05:05] <agronholm> but for some reason, when I start iterating entry points, it errors out because I don't have that package installed on 3.5

[00:07:39] <agronholm> it's like setuptools is ignoring the conditionals

[00:15:31] <agronholm> is automatic wheel building supposed to work on 3.5.0rc4?

[00:17:44] <agronholm> hrm, upgrading to rc4 seemed to fix that entry point problem

[00:20:55] <agronholm> gah, no

[01:02:20] <agronholm> alright, I figured it out -- pkg_resources was picking up the egg-info from my project dir

[01:03:35] <agronholm> quite a dilemma

[01:03:57] <agronholm> I have the distribution installed in develop mode, which writes its .egg-info in the project directory

[01:05:03] <agronholm> then I run tox and iterate over the project's entry points, which fails because the pkg-info lists the dependencies for the python version I use for development

[01:20:14] <agronholm> this is what my requires.txt in the egg-info directory looks like: https://bpaste.net/show/5ba5504bef15

[01:20:20] <agronholm> why is "typing" there twice?

[06:32:16] <ronny> agronholm: whats in setup.py?

[06:32:47] <ronny> agronholm: also note that a install_requires on setuptools is kinda pointless for most things ^^

[13:41:37] <agronholm> ronny: is it? I've been told before by downstream packagers (debian) that it needs to be there so they can determine that it depends on pkg_resources

[13:43:04] <agronholm> OTOH, will any distro ship python 3.4 without setuptools?

[13:45:52] <agronholm> for reference, setup.py: https://github.com/asphalt-framework/asphalt/blob/master/setup.py

[14:37:10] <ionelmc> agronholm: but do you need it at runtime?

[14:37:22] <agronholm> ionelmc: yes

[14:37:31] <ionelmc> agronholm: what for

[14:37:32] <agronholm> I use pkg_resources

[14:37:41] <agronholm> for loading entry points

[14:37:45] <ronny> agronholm: i suppose then its still sensible to have around

[14:37:49] <agronholm> yeh

[14:38:24] <ionelmc> agronholm: that's not a hard runtime dependency

[14:38:33] <agronholm> ionelmc: eh?

[14:38:49] <ionelmc> pip creates scripts that just import whatever you declared in the entrypoint, no pkg_resources needed

[14:39:03] <agronholm> ionelmc: how will that help me load plugins?

[14:39:12] <ionelmc> oh

[14:39:20] <ionelmc> i thought you only used console_scripts :)

[14:39:25] <agronholm> don't assume :)

[14:40:14] <ionelmc> reddit distortion, there was a similar q a day ago

[14:41:09] <agronholm> but what I don't get is why typing shows up in both unconditional requirements and conditional requirements

[14:41:16] <agronholm> since pkg_resources can happily use both

[14:41:17] <ionelmc> one person even suggested to use scripts instead of console_scrpts entrypoints, yuk ...

[14:44:49] <ronny> agronholm: i recall pytest-cv has a very specific coverage requirement, installing a newer version of coverage breaks its world

[14:45:02] <ronny> ionelmc: can that be eleevated?

[14:45:07] <ronny> *pytest-cov

[14:45:24] <agronholm> ronny: ok, but how is that relevant?

[14:45:25] <ionelmc> wat

[14:45:37] <ionelmc> pytest-cov supports coverage 4.0 since 2.0.0

[14:45:43] <ronny> agronholm: versionconflict happesn when the working set has a conflict

[14:45:48] <ronny> ionelmc: then maybe thats the issue

[14:45:50] <ionelmc> 2.1.0 added support for 4.0b3

[14:46:01] <ronny> agronholm: can you fgure which package adds the conflict?

[14:46:25] <agronholm> I'm confused

[14:46:26] <ionelmc> what conflict?

[14:46:47] <ionelmc> you mean leftover scripts created by easy_install (that have version pin in them) ?

[14:47:31] <agronholm> what are we even talking about here?

[14:47:46] <agronholm> I'm talking about requires.txt in the generated egg-info

[14:48:44] <ionelmc> to be fair it's hard to know what you're talking about when you're talking about python packaging

[14:48:46] <ionelmc> xD

[14:49:21] <agronholm> so I'm wondering why "setup.py egg_info" adds typing twice to .egg_info/requires.txt, given my setup.py

[14:50:01] <ionelmc> agronholm: cause it's wrong :)

[14:50:10] <ionelmc> you list that dep twice

[14:50:14] <agronholm> I don't?

[14:50:16] <ionelmc> both in install_requires and extras

[14:50:29] <agronholm> hrm

[14:50:53] <ionelmc> it is completely wrong to put conditional dep in instsall_requires

[14:51:04] <ionelmc> universal wheels will build improperly

[14:51:17] <agronholm> unless your conditions are complex...though mine here is not

[14:51:38] <ionelmc> i assume you gonna use univ wheels since you don't have any c ext

[14:51:55] <agronholm> I have the specifiers set in setup.cfg

[14:52:03] <ionelmc> plus it's not trivial to make non-universal wheels when you don't have c exts

[14:52:26] <agronholm> I didn't think extras_require would work properly for sdists

[14:52:30] <agronholm> I guess I was wrong

[14:52:56] <ionelmc> ps

[14:53:20] <ionelmc> your tag also looks odd

[14:53:27] <agronholm> how so

[14:53:34] <ionelmc> why just py34.py35 ?

[14:53:43] <agronholm> because it requires at least py34

[14:53:58] <agronholm> and there seems to be no other way to indicate the minimum python version

[14:54:04] <ionelmc> i have bad feeling about it :)

[14:54:09] <agronholm> why

[14:54:19] <ionelmc> what do you do for 3.6 ?

[14:54:41] <agronholm> see if the idiotic metadata mess has been fixed by then, or just add .py36

[14:55:01] <ionelmc> prolly irrelevant cause you'll make new releases with correct tag by the time 3.6 is released for just a thought

[14:55:09] <agronholm> I have though about it

[14:55:22] <ionelmc> s/for/but/

[14:55:23] <agronholm> iirc there used to be a way to specify the minimum python version in setup.cfg

[14:55:41] <ionelmc> dunno

[14:55:45] <agronholm> but the setup.cfg metadata was "deprecated" without any replacement being provided

[14:55:46] <ionelmc> i woulnd't bother

[14:56:08] <ionelmc> raising an error at runtime if python version is not good seems like the right way

[14:56:09] <agronholm> for one, you used to be able to specify things like "install this only for python < X.Y"

[14:56:19] <agronholm> but that doesn't work for wheels

[14:56:30] <ionelmc> run time

[14:56:34] <agronholm> unless you mean in the actual code

[14:56:34] <ionelmc> when your code runs

[14:56:41] <ionelmc> eg: when someone tries to import it

[14:56:51] <agronholm> eh, add checks to every single module?

[14:57:05] <ionelmc> no, just the main package __init__.py

[14:57:11] <agronholm> right

[14:57:17] <ionelmc> cause that gets run regardless of what you import

[14:57:20] <agronholm> but I think I have my bases covered with the check in setup.py and the wheel tags

[14:57:56] <ionelmc> check too early - suffer too soon

[14:58:02] <agronholm> ?

[14:58:16] <ionelmc> maybe it's just a matter of style

[14:58:58] <ionelmc> install of packages in python has always been very loose

[14:59:04] <agronholm> yeah :/

[14:59:08] <ionelmc> so i woulnd't bother with constraints there

[14:59:12] <agronholm> well, that fixed my actual problem now

[14:59:13] <ionelmc> cause it's pointless

[14:59:37] <agronholm> and the tox run on py34 passed

[14:59:48] <ionelmc> iow: some user might actually success to install your package on, say, python2

[14:59:54] <ionelmc> so you still need a runtime check

[15:00:04] <ionelmc> so why not only do it at runtime

[15:00:08] <ionelmc> that was my thought

[15:00:14] <agronholm> they would have to go out of their way to do so

[15:00:27] <agronholm> not use pip install or easy_install or even setup.py install

[15:01:02] <ionelmc> what if they make egg

[15:01:08] <agronholm> how could they

[15:01:10] <ionelmc> :>

[15:01:12] <ionelmc> no idea

[15:01:21] <agronholm> I mean, that requires you to run setup.py

[15:02:18] <ionelmc> they could make bdist_dumb on 3.4 and put that somewhere in py2.7's site-pakcages i suppose

[15:02:19] <agronholm> the point of these checks is to prevent idiots from filing bugs when the distro won't work in the wrong python version

[15:02:48] <agronholm> ionelmc: anyone knowledgeable enough to do that is smart enough to figure this out :P

[15:05:26] <agronholm> ionelmc: thanks for the input, the problem is now solved

[18:24:10] <ionelmc> this is intreresting, seems pkg_resources don't return entry point at all if it's from a package with unresolved deps

[18:24:13] <ionelmc> is this normal?

[19:13:18] <ronny> ionelmc: well, its a possible outcome for a multiversion setup

Log file Viewer

Help | Karma | Search:

#pypa logs for Saturday the 12th of September, 2015