专栏首页Frost's BlogA deeper look into Pipenv and Poetry

A deeper look into Pipenv and Poetry

It is 8 months passed since I posted the article comparing Pipenv with Poetry, which is the most popular article in my blog now. However, it was not a good review of the two tools, I have not even read the documentation of Poetry. In the end of last year I became a collaborator of Pipenv and util then have I realized there are so many trade-offs and, well, defects in Pipenv. In the area of software engineering, the successor always wins. The creator can't anticipate all corner cases in his prototype or original thoughts, especially for such a CLI tool that are run on millions of computers with totally different environment setup.

Main problems with Pipenv

  • Introduced more files of a new format, which is not perfect also. See how many miscellaneous files there are under the project root:
  • The commands and options are always confusing and unclear, and the options are not fully tested to ensure the availability.
  • Locking performance, there are many issues about dependency resolution. The rollment of passa may bring a big improvement, but the timeline is still not decided yet.
  • Regression issues burned out users' patience. The test coverage is still very low, though testing against cross-platform and multi environment is a difficult and complicated mission.
  • The maintainers are maintaining a large amount of projects backing Pipenv which need to be updated at a fairly high frequency. And it in turn brings high risk of regression. All PEEPs(Pipenv's enhancement proposals) need BDFL's approval while Kenneth Reitz is bothered by mental healthy and takes the back seat for a long time.

I myself are using Pipenv in daily work and help maintain Pipenv, too. So I wish a better future for Pipenv.

What about Poetry

Poetry seems a better choice ya? I am also considering transfering to Poetry, but before that, I have to point out some downsides of it.

  • Poetry doesn't work well with PyPA's current packaging system. It is very common that people clone a VCS repository and work on it. However, you can't pip install a Poetry project without extra steps to generate a setup.py from pyproject.toml.
  • It is weird that Poetry puts python requires in dependency section. Python requires should be placed in global settings, although it shares the same version specifiers with normal dependencies.
  • Neither Pipenv or Poetry supports to activate a virtualenv outside of project directory. I know it is inspired by NPM but in Python world people are likely to put some scripts in places other than the project directory. It is supported by virtualenvwrapper.
  • Poetry only works under one workflow. For example, it doesn't support installing current dependencies into system Python, which is the typical workflow for developing in docker.

What I expect in the future of Python packaging

I like the idea of "Every project is a package" and the introduction of pyproject.toml. It is introduced in PEP-508 but it is not finalized yet.

After PEP-508 is finalized, it is better that there is an official way to define dependencies in pyproject.toml, not in section like tool.poetry. It may look like:

Toml

[project]
name = "poetry"
version = "0.12.10"
description = "Python dependency management and packaging made easy."
authors = [
    "Sébastien Eustace <sebastien@eustace.io>"
]
license = "MIT"
readme = "README.md"
python = "~2.7 || ^3.4"
homepage = "https://poetry.eustace.io/"
repository = "https://github.com/sdispater/poetry"
documentation = "https://poetry.eustace.io/docs"

keywords = ["packaging", "dependency", "poetry"]

classifiers = [
    "Topic :: Software Development :: Build Tools",
    "Topic :: Software Development :: Libraries :: Python Modules"
]

# Requirements
[dependencies]
cleo = "^0.6.7"
requests = "^2.18"
cachy = "^0.2"
requests-toolbelt = "^0.8.0"
jsonschema = "^3.0a3"
pyrsistent = "^0.14.2"
pyparsing = "^2.2"
cachecontrol = { version = "^0.12.4", extras = ["filecache"] }
pkginfo = "^1.4"
html5lib = "^1.0"
shellingham = "^1.1"
tomlkit = "^0.5.1"

# The typing module is not in the stdlib in Python 2.7 and 3.4
typing = { version = "^3.6", python = "~2.7 || ~3.4" }

# Use pathlib2 for Python 2.7 and 3.4
pathlib2 = { version = "^2.3", python = "~2.7 || ~3.4" }
# Use virtualenv for Python 2.7 since venv does not exist
virtualenv = { version = "^16.0", python = "~2.7" }
# functools32 is needed for Python 2.7
functools32 = { version = "^3.2.3", python = "~2.7" }

[dev-dependencies]
pytest = "^3.4"
pytest-cov = "^2.5"
mkdocs = { version = "^1.0", python = "~2.7.9 || ^3.4" }
pymdown-extensions = "^6.0"
pygments = "^2.2"
pytest-mock = "^1.9"
pygments-github-lexers = "^0.0.5"
black = { version = "^18.3-alpha.0", python = "^3.6" }
pre-commit = "^1.10"
tox = "^3.0"
pytest-sugar = "^0.9.2"

[scripts]
poetry = "poetry.console:main"

I hope pyproject.toml will eventually replace setup.py, and in the transition period, Pip, or whatever name, should be able to read both files. The "Pip" should be a combination of Pipenv and Poetry and be the ultimate solution for Python packaging.


Update

Another drawbacks I found recently: Pipenv, or more precisely, virtualenv, doens't use the built-in venv module to create a VE, which may bring troubles:

  • When the Python interpreter is updated, the VE will become stale.
  • Some packages such as matplotlib requires framework build on macOS. Virtualenv creates non-framework build even if the original one is framework-build

I forked the virtualenv with a patch to address these problems. Replace the PyPI virtualenv via:

Bash

$ pip install -I https://github.com/frostming/virtualenv-venv/releases/download/16.2.0-fork/virtualenv-16.2.0_fork-py2.py3-none-any.whl

Everything works well now.

本文参与腾讯云自媒体分享计划,欢迎正在阅读的你也加入,一起分享。

我来说两句

0 条评论
登录 后参与评论

相关文章

  • What's the problem about Python packaging?

    I have been using Python as my primary programming language for 4 years. It is e...

    岂不美哉Frost
  • 在博客与笔记中使用Markdown

    这些我都不是特别满意,于是便萌生了干脆建一个个人博客的想法。通过多方考查,得到一个比较好的解决方案:Hexo + Github page。Github page...

    岂不美哉Frost
  • 使用 Flask 做一个评论系统

    因为我博客使用的Disqus代理服务下线,博客的评论系统可能有一阵子没有工作了。惭愧的是我竟然最近才发现,我的工作环境一直是没有GFW存在的,发现是因为有个朋友...

    岂不美哉Frost
  • 基于不确定性原理的优化;新的元启发式策略框架(CS NE)

    为了更灵活地在勘探与开发之间取得平衡,本文提出了一种基于不确定性原理的新的元启发式方法。事实证明,UP在科学的多个分支中都是有效的。在量子力学的分支中,在任何量...

    小童
  • 在SAP HANA Express Edition里创建数据库表

    在默认的HXE database tenant上启用document store:

    Jerry Wang
  • 在SAP HANA Express Edition里创建数据库表

    在默认的HXE database tenant上启用document store: ALTER DATABASE HXE ADD ‘docstore’;

    Jerry Wang
  • 功能和功能需求的实现和执行建模(CS SE)

    需求工程在软件系统的开发中扮演着重要的角色。 这个过程中最困难的任务之一是识别功能需求。 许多项目中的一个关键问题是直到开发周期的后期才有需求。 本文研究的核心...

    用户7095611
  • Job Prospects of AI

    Job Prospects of AI Machine learning engineer is becoming a job, whose demand ga...

    企鹅号小编
  • 对功能和功能需求的实现和执行进行建模(CS.SE)

    需求工程在开发软件系统中起着至关重要的作用。在此过程中最困难的任务之一就是确定功能需求。在许多项目中,一个关键问题是直到开发周期的后期才缺少需求。在本文中,我们...

    蔡小雪7100294
  • 【论文推荐】最新6篇目标检测​(Object Detection)相关论文—物体链接、手机端、三维地图、航空图像、检测与姿态估计

    【导读】专知内容组整理了最近六篇目标检测(Object Detection)相关文章,为大家进行介绍,欢迎查看! 1. Object Detection in ...

    WZEARW

扫码关注云+社区

领取腾讯云代金券