second take at setuptools_scm #1834

RonnyPfannschmidt · 2016-08-21T18:01:03Z

since setuptools 18.6 fixes the issues with develop installs

https://github.com/pypa/setuptools/blob/master/CHANGES.rst#186

coveralls · 2016-08-21T18:10:58Z

Coverage decreased (-0.02%) to 93.0% when pulling de0f0cf on RonnyPfannschmidt:setuptools-scm-take-2 into abe8f5e on pytest-dev:master.

nicoddemus · 2016-08-22T17:12:37Z

HOWTORELEASE.rst

@@ -3,11 +3,9 @@ How to release pytest

 Note: this assumes you have already registered on pypi.

-1. Bump version numbers in ``_pytest/__init__.py`` (``setup.py`` reads it).
+1. Check and finalize ``CHANGELOG.rst``.


We probably want to mention to generate the tag locally right before creating the package, correct?

im considering a setuptools_scm feature where the branch name would be considered as well

nicoddemus · 2016-08-22T17:13:10Z

Besides my comment on HOWTORELEASE, linting has fail due to the fact that the numbers in the sections don't increase sequentially anymore.

nicoddemus · 2016-08-29T13:58:18Z

My previous comment is still valid, I think:

We probably want to mention to generate the tag locally right before creating the package, correct?

😁

RonnyPfannschmidt · 2016-08-29T14:06:04Z

actually i have a pending setuptools_scm update for solving those details more nice - will show later today

coveralls · 2016-08-29T14:23:43Z

Coverage decreased (-0.02%) to 93.008% when pulling d6e1021 on RonnyPfannschmidt:setuptools-scm-take-2 into e75078e on pytest-dev:master.

nicoddemus · 2017-04-19T11:43:26Z

Did you give up on this, or will try on another branch?

RonnyPfannschmidt · 2017-04-19T12:39:24Z

@nicoddemus i'll pick this up when either setuptools_scm supports branches or we decide to get rid of the features branch

nicoddemus · 2017-04-19T12:41:38Z

I might be missing something, but I don't really see why. Doesn't setuptools_scm determine the version based on the nearest tag? So if I tag a commit 2.1.1 the version will be 2.1.1, and if I decide to create a new feature release I can tag a commit 2.2.0 and it will just work, right?

RonnyPfannschmidt · 2017-04-19T12:58:01Z

@nicoddemus yes, but its a bit of a pain when a feature branch for example has a "wrong" version attached for quite a while

nicoddemus · 2017-04-19T13:01:53Z

Can you elaborate a bit what do you mean by "its a bit of a pain when a feature branch for example has a "wrong" version attached for quite a while"?

I insist because I would love to use setuptools_scm on pytest, and would like to understand the reason completely if we have to ditch it because of having a features branch.

RonnyPfannschmidt · 2017-04-19T13:09:18Z

@nicoddemus setuptools_scm currently works based only on tags, so its not aware on how branches should bump the last version, and while it is supported to put certain tags in to help, its not exactly well received to litter the history with extra tags to keep setuptools_scm happy

nicoddemus · 2017-04-19T13:13:06Z

Sorry, I still don't quite get what the issue is, bear with me... from my POV the version is important basically only when making a release, and at that point we will have a tag at the exact location we need it to be. Or am I I missing something?

RonnyPfannschmidt · 2017-04-19T13:14:07Z

if we are fine with such a limitation, we can just switch over right now

nicoddemus · 2017-04-19T13:19:04Z

I'm 👍 on it, unless I'm missing a situation where the development version is important.

RonnyPfannschmidt · 2017-04-19T14:09:16Z

will rebase after pass

coveralls · 2017-04-19T14:57:36Z

Coverage decreased (-0.02%) to 92.614% when pulling 6a691b3 on RonnyPfannschmidt:setuptools-scm-take-2 into 80cabca on pytest-dev:features.

nicoddemus · 2017-04-19T15:29:59Z

👍

Linting broke because of some bad interaction between setuptools-scm and check-manifest...

since setuptools 18.6 fixes the issues with develop installs https://github.com/pypa/setuptools/blob/master/CHANGES.rst#186 pypa/setuptools#439

nicoddemus · 2017-04-20T15:45:16Z

can we include egg-info into frozen pytest

Unfortunately no. It think all freeze tools don't support including the packaging metadata (see marcelotduarte/cx_Freeze#216) and I suspect even after including package metadata pkg_resources would still need to learn how to deal with that metadata embedded in zipfiles.

note that using unknown as version would break expectations as well

I understand, but None breaks pytest completely... "unknown" as version would break some expectations, but I think it preferable than the former.

approach breaks check-manifest

I would prefer if we found a solution to that, given that check-manifest is a tool we use internally, as opposed to freezing which is something normal users use.

Just to add some context, we freeze pytest extensively together with our applications at work, so we can run the tests using the frozen executable. This is invaluable to catch packaging errors.

coveralls · 2017-04-20T20:11:46Z

Coverage decreased (-0.02%) to 92.614% when pulling 4242bf6 on RonnyPfannschmidt:setuptools-scm-take-2 into 80cabca on pytest-dev:features.

nicoddemus · 2017-04-21T03:36:42Z

tox.ini

@@ -47,6 +47,9 @@ commands=

 [testenv:linting]
 basepython = python2.7
+# needed to keep check-manifest working
+setenv =
+    SETUPTOOLS_SCM_PRETEND_VERSION=2.0.1


So it seems this fixes the check-manifest issue. 😁

Can we switch to the write_to option now?

RonnyPfannschmidt · 2017-04-21T07:17:45Z

Nppe, since write_to breaks check manifest

nicoddemus · 2017-04-22T20:10:49Z

Nppe, since write_to breaks check manifest

I'm a little confused then, why do you need to set SETUPTOOLS_SCM_PRETEND_VERSION during linting?

RonnyPfannschmidt · 2017-04-22T20:15:09Z

@nicoddemus pretend_version is for the breakage that is incurred from not having version data in whatever check-manifest copies around, the other breakage is incurred because check-manifest does nit allow for generated files in a sdist that aren't in the scm

nicoddemus · 2017-04-22T20:22:03Z

the other breakage is incurred because check-manifest does nit allow for generated files in a sdist that aren't in the scm

Perhaps we can use the ignore option to ignore the generated version file?

[check-manifest]
ignore =
    .travis.yml

https://pypi.python.org/pypi/check-manifest/

RonnyPfannschmidt · 2017-04-23T14:59:29Z

@nicoddemus thanks for that hint

RonnyPfannschmidt · 2017-04-26T06:50:56Z

@nicoddemus unknown breaks hypothesis - im going to go back to None

nicoddemus · 2017-04-26T12:38:26Z

Hmm wait, but this was working before with hypthosis right? I mean, hypothesis should be seeing the correct version obtained from _version.py.

But if we do find a reasonable explanation for the above, how about using 0.0.0 then? That's a valid version string but it is clear something is incorrect with the installation. None is not better because pytest itself breaks if version is None.

RonnyPfannschmidt · 2017-04-26T13:40:11Z

pytest i can directly fix, hypothesis i cant

nicoddemus · 2017-04-26T13:52:08Z

Why you don't like my suggestion of "0.0.0"? It is as good as "unknown" or None I think.

RonnyPfannschmidt · 2017-04-26T14:23:15Z

@nicoddemus unknown doesnt break on bad version checks due to lexical order, 0.0.0 does in any case
i'd rather have a correct failure than trying to keep stuff working by means of pretense

coveralls · 2017-04-26T14:32:39Z

Coverage decreased (-0.1%) to 92.505% when pulling a280e43 on RonnyPfannschmidt:setuptools-scm-take-2 into 80cabca on pytest-dev:features.

nicoddemus · 2017-04-26T14:34:34Z

i'd rather have a correct failure than trying to keep stuff working by means of pretense

Fair enough. 👍

RonnyPfannschmidt · 2017-04-26T15:18:07Z

@nicoddemus as far as i can tell introducing setuptools_scm will warrant a major release due to the cahnge to pytest.__version__ behaviour

The-Compiler · 2017-04-26T15:23:22Z

What changes about it?

RonnyPfannschmidt · 2017-04-27T09:17:47Z

it could be none, and it an include local version components,
as such it will break naive version parsers

nicoddemus · 2017-04-27T12:08:48Z

it could be none, and it an include local version components,
as such it will break naive version parsers

But that doesn't affect releases, which will always have a __version__ in the same format as always. It doesn't make sense to me to increase the major version because of a change that doesn't affect the releases at all. In other words, from a user's POV a release 3.1.0 with or without this change is indistinguishable from each other.

Do you agree?

nicoddemus · 2017-04-27T12:10:20Z

Merging as this now looks great, thanks @RonnyPfannschmidt!

The-Compiler · 2017-05-09T20:41:25Z

This breaks hypothesis, and maybe other plugins doing pytest version checking: HypothesisWorks/hypothesis#611

I still maintain my opinion that the benefits of saving some 10s of work for bumping the version number (which can trivially be automated, too) is by no means worth the breakage this causes everywhere...

nicoddemus · 2017-05-09T21:54:51Z

is by no means worth the breakage this causes everywhere...

Definitely I agree, but that was not supposed to happen. Can you provide more details on how it was possible to install pytest and somehow setuptools_scm failing to generate a _version.py file? Puzzling...

The-Compiler · 2017-05-10T04:59:49Z

Ah, doing another pip install -e run (after pulling the setuptools-scm change) fixes it - sorry for not trying this before. This was a develop install done before setuptools-scm was introduced.

I wanted to run a pytest --help to check something before going to bed and then got this 😆

~~Now I'm curious though - what's this, and why 1.6.0?~~

edit: Oh. (pytest-mock) version, not pytest (mock-version). 😆

$ cat .venv/lib/python3.6/site-packages/_pytest_mock_version.py
# coding: utf-8
# file generated by setuptools_scm
# don't change, don't track in version control
version = '1.6.0'

nicoddemus · 2017-05-10T12:38:12Z

Ah, doing another pip install -e run (after pulling the setuptools-scm change) fixes it - sorry for not trying this before. This was a develop install done before setuptools-scm was introduced.

I figured, no worries. This will only happen if one has an old installation and didn't execute a pip install -e. Released packages will always have a _version.py file so this should not be an issue. 😉

RonnyPfannschmidt added the type: proposal proposal for a new feature, often to gather opinions or design the API around the new feature label Aug 21, 2016

nicoddemus reviewed Aug 22, 2016
View reviewed changes

RonnyPfannschmidt force-pushed the setuptools-scm-take-2 branch from 5b7113f to d6e1021 Compare August 29, 2016 13:56

RonnyPfannschmidt changed the title ~~second take at setuptools_scm~~ [wip] second take at setuptools_scm Aug 29, 2016

RonnyPfannschmidt closed this Apr 19, 2017

RonnyPfannschmidt reopened this Apr 19, 2017

RonnyPfannschmidt changed the base branch from master to features April 19, 2017 14:07

RonnyPfannschmidt added 4 commits April 19, 2017 19:40

second take at setuptools_scm

c3aee4b

since setuptools 18.6 fixes the issues with develop installs https://github.com/pypa/setuptools/blob/master/CHANGES.rst#186 pypa/setuptools#439

HOWTORELEASE.tst: use restructuredtext autonumbering

31e6fe8

restore check-manifst functionality

c0a51f5

restore linting, drop _pytest._version for check-manifest

2cf4227

RonnyPfannschmidt force-pushed the setuptools-scm-take-2 branch from 6a691b3 to 2cf4227 Compare April 19, 2017 18:26

RonnyPfannschmidt changed the title ~~[wip] second take at setuptools_scm~~ second take at setuptools_scm Apr 19, 2017

use unknown to specify unknown versions

4242bf6

nicoddemus reviewed Apr 21, 2017

View reviewed changes

restore setuptools_scm write_to usage

e02cb6d

fix import error

a280e43

nicoddemus merged commit cccb2cc into pytest-dev:features Apr 27, 2017

The-Compiler mentioned this pull request May 9, 2017

pytest's switch to setuptools-scm breaks version check HypothesisWorks/hypothesis#611

Closed

RonnyPfannschmidt deleted the setuptools-scm-take-2 branch June 11, 2017 13:43

second take at setuptools_scm #1834

second take at setuptools_scm #1834

Conversation

RonnyPfannschmidt commented Aug 21, 2016

coveralls commented Aug 21, 2016

nicoddemus Aug 22, 2016

Choose a reason for hiding this comment

RonnyPfannschmidt Aug 23, 2016

Choose a reason for hiding this comment

nicoddemus commented Aug 22, 2016

nicoddemus commented Aug 29, 2016

RonnyPfannschmidt commented Aug 29, 2016

coveralls commented Aug 29, 2016

nicoddemus commented Apr 19, 2017

RonnyPfannschmidt commented Apr 19, 2017

nicoddemus commented Apr 19, 2017 • edited Loading

RonnyPfannschmidt commented Apr 19, 2017

nicoddemus commented Apr 19, 2017

RonnyPfannschmidt commented Apr 19, 2017

nicoddemus commented Apr 19, 2017

RonnyPfannschmidt commented Apr 19, 2017

nicoddemus commented Apr 19, 2017

RonnyPfannschmidt commented Apr 19, 2017

coveralls commented Apr 19, 2017

nicoddemus commented Apr 19, 2017

nicoddemus commented Apr 20, 2017

coveralls commented Apr 20, 2017

nicoddemus Apr 21, 2017

Choose a reason for hiding this comment

RonnyPfannschmidt commented Apr 21, 2017

nicoddemus commented Apr 22, 2017

RonnyPfannschmidt commented Apr 22, 2017

nicoddemus commented Apr 22, 2017 • edited Loading

RonnyPfannschmidt commented Apr 23, 2017

RonnyPfannschmidt commented Apr 26, 2017

nicoddemus commented Apr 26, 2017

RonnyPfannschmidt commented Apr 26, 2017

nicoddemus commented Apr 26, 2017

RonnyPfannschmidt commented Apr 26, 2017

coveralls commented Apr 26, 2017

nicoddemus commented Apr 26, 2017

RonnyPfannschmidt commented Apr 26, 2017 • edited by nicoddemus Loading

The-Compiler commented Apr 26, 2017

RonnyPfannschmidt commented Apr 27, 2017

nicoddemus commented Apr 27, 2017

nicoddemus commented Apr 27, 2017

The-Compiler commented May 9, 2017

nicoddemus commented May 9, 2017

The-Compiler commented May 10, 2017 • edited Loading

nicoddemus commented May 10, 2017

nicoddemus commented Apr 19, 2017 •

edited

Loading

nicoddemus commented Apr 22, 2017 •

edited

Loading

RonnyPfannschmidt commented Apr 26, 2017 •

edited by nicoddemus

Loading

The-Compiler commented May 10, 2017 •

edited

Loading