I was looking to package my project, Ciphey, for operating systems and for managers that aren’t PyPi. Unfortunately, there seemed to be very little information on the web about this.
This is a guide on packaging your Python project for:
- Windows Package Manager
- Arch User Repository
Semantic Versioning is a system defining how to write version numbers. The 3 numbers are:
If you have fixed some bugs, increment the bugs counter.
If you have added a minor feature, increment the minor counter.
If you have done something major, increment the major counter.
We can signify whether a release is still being rested or not by adding “rc” (release candidate) to the end of the version. “5.0.0rc1” signifies “release candidate 1” which means this is the first public testing release of version 5.0.0.
The traditional method of using Setuptools is outdated and old. There’s a new cowboy in town, and their name is Poetry.
Poetry creates a
pyproject.yml file, which is the successor to the 3-file Setuptools system.
You should use Poetry because:
- Instead of 3 files, it is now 1 file
- That 1 file is yaml, which means you don’t have to cry reading weirdly formatted Python code used as a replacement for yaml
- You can separate dev & normal dependencies, so users only install the dependencies they need to run the app
- It’s physically easier to build and upload to PyPi
- And a host of other things
Install poetry with
pip3 install poetry.
Navigate to your projects directory and run the command
This will generate a
pyproject.toml file. This file contains everything PyPi needs to upload your project and allow other users to download it.
Let’s run through the sections in this new file, and fill them out as we go.
This section relates to the metadata of your project.
The name, description, version etc.
To add a new variable to a .toml file, write it as if it were python
name = "project name".
The most important ones we’ll want are:
What is the name of the project?
name = "Ciphey`
The semantic version number.
version = "5.0.0rc6"
Short one-line description.
description = "Automated decryption tool"
A list containing the author's name and email address. In the format of
authors = ["brandon <firstname.lastname@example.org>"]
What license does the project use?
license = "MIT"
The name of the
README.md file, in the same directory, as
readme = "README.md"
Here is what Ciphey’s Pyproject looks like:
[tool.poetry] name = "ciphey" version = "5.0.0rc4" description = "Automated Decryption Tool" authors = ["Brandon <brandon@skerritt (dot) blog>"] license = "MIT" documentation = "https://docs.ciphey.online" exclude = ["tests/hansard.txt"] readme = "README.md"
To find out what else you can include, check out the Poetry documentation.
A list of all the dependencies your project uses. Don’t worry! You don’t have to manually add them to the list.
Run the command
poetry add <dependency> to add a dependency to your project, and Poetry will automatically add it to your project.
For example, to add
rich we would write
poetry add rich.
The dependencies the developers of your tool rely on. To add the testing library Pytest, write
poetry add --dev pytest.
This is the 2nd most important part! As it contains an entry point. When a user installs your package, you’d probably like them to be able to run
package in their terminal and use your package.
For a folder called Ciphey with a file named Ciphey.py with a function called main(), we would write:
ciphey = 'ciphey.ciphey:main'
When the user runs the command
ciphey, it actually runs the main function of the file ciphey of the folder ciphey.
For a more detailed explanation, check this out.
poetry run runs our script but includes the
pyproject.toml file. It executes the given command inside of a virtual environment, essentially it allows us to test our project in the same way a user might run our program.
Poetry install installs our program and all its dependencies. Essentially, we can test exactly what our users will do when they install the project for themselves.
poetry update updates our dependencies.
Now onto the most important commands.
Poetry build builds our project, which means it generates some files that other users can install and use on their system.
Poetry Publish will publish the package to PyPi for us. All we need to do is enter a username/password.
And just like that, our project is now on PyPi for the whole world to download!
Windows Package Manager (WinGet)
There are 3 steps to submitting your app to WinGet.
- Turning your project into an EXE
- Create a manifest file
- Create a pull request on GitHub
Turning your project into an EXE
Windows requires an exe file for Python projects. Don’t fret! Turning your project into an exe file isn’t that difficult.
Download Pyinstaller, and then create an
entry point in the root directory.
An entry point is a Python file that when run, will run the program. So for instance:
Ciphey/ |-entry_point.py |-ciphey/ |----main.py
Here the project code is in
ciphey/. To run the program, we’d have to run
This is the typical file structure of a project. We have README.md and similar in the root, and all the source code in a sub-directory.
We must create a program that calls our program using an entry point. The entry point is a function which runs the program.
If you have a
pyproject.yml, you may already have entry points defined in them.
entry_point.py, we’d write:
from ciphey.__main__ import main if __name__ == '__main__': main()
Or however, your entry point is defined.
Now, run Pyinstaller on the
entry_point.py file. It will generate a new folder called
dist/. If you are using the default Python
.gitignore it will automatically disclude
dist/. If not, please add it to the
Pyinstaller will also generate a
entry_point.spec file in the root directory. This file contains a specification of how Pyinstaller should build your program.
Once built, run the program.
You will likely see some errors. Most likely bad imports. To fix this, define them in the spec file. There are a couple main lists in the spec file you should know about:
If you already hold a binary for a specific file/library (such as C++ compiled) then define the location here.
If your program relies on .txt data or anything that isn’t a Python file define it here.
- Hidden Imports
Hidden imports look like
__import__. These types of imports are not visible to Pyinstaller’s analysis, so to get around this define the imports in the Hidden imports section.
Exclude a package from Pyinstaller. I had an error with Setuptools, but since Setuptools isn’t needed to run Ciphey I told Pyinstaller to exclude the entire package.
If you have edited the
.spec file, you may want to delete the line the
.gitignore which states to ignore the file. Otherwise, all the hard work you’ve spent in getting Pyinstaller to work may be lost.
One thing to note. Pyinstaller only creates executables for the system you are on. You cannot build a Windows .exe from Linux. You have to be on the system you are trying to build. Pyinstaller can create these binaries (assuming you own the operating systems for them):
- Windows .exe
- Mac OS’ .app
- Linux’s Binary Files
Once built and packaged, running the executable on Windows / Mac will bring up the Terminal (assuming you do not have any GUI). If you do, it will bring up the GUI.
Pyinstaller packages an entire Python interpreter with the binary, So do not worry about the user having Python installed.
Now we have an exe. I would suggest converting to MSIX.
Exes are cool and all, but they’re not really package manager material. Winget with a .exe is a glorified Softonic / CNET Software. MSIX gives us the real power to:
- Automatically update the app
- Uninstall cleanly
- Understand what dependencies are required
So in short, if we use a .exe it is just a glorified Software downloader. If we use .msix it is more of a package manager.
Below is the
manifest.yml file for Ciphey.
Id: Ciphey.Ciphey Publisher: Ciphey Name: Ciphey Version: 5 AppMoniker: Ciphey MinOSVersion: 10.0.0.0 Description: Automated Decryption Tool Homepage: https://www.github.com/ciphey/ciphey License: MIT LicenseUrl: https://opensource.org/licenses/MIT InstallerType: exe Installers: - Arch: x64 Url: https://statics.teams.cdn.office.net/production-windows-x64/1.3.00.4461/Teams_windows_x64.exe Sha256: 712f139d71e56bfb306e4a7b739b0e1109abb662dfa164192a5cfd6adb24a4e1 ManifestVersion: 0.1.0
Ciphey’s manifest file Let’s walk through quickly what each potentially confusing part is. I’ll leave things like “Name” and “License” as they are self-explanatory.
This follows a
publisher.package format. Your publisher's name, and then the package name.
- App Moniker
What’s the common name people search for your package with? For example “Visual Studio Code” can also be found with “vscode”.
If you have a long package name and your users often shorten it, this will be helpful to you.
What is the minimum version of Windows the app supports?
10.0.0.0 means Windows 10 and above.
What type of installer do you have? If you followed along, you should have a .exe file.
This details how to install your package.
What architecture does your project support? Examples include
arm, arm64, x86, x64 and neutral.
The ones you’re probably after are either x86 or x64. x86 is for 32-bit operating systems, x64 is for 64-bit operating systems.
What is the URL of the .exe installer for your project? Personally, I hosted this on GitHub.
What is the SHA256 checksum of your .exe file? There’s an article here detailing how to calculate it.
- Manifest Version
Every time you update this file, update the manifest version numbering using Semantic Versioning.