What happens when (your?) package doesn't work with PyInstaller? Say you have data files that you need at runtime? PyInstaller doesn't bundle those. Your package requires others which PyInstaller can't see? How do you fix that?
In summary, a "hook" file extends PyInstaller to adapt it to the special needs and methods used by a Python package. The word "hook" is used for two kinds of files. A runtime hook helps the bootloader to launch an app, setting up the environment. A package hook (there are several types of those) tells PyInstaller what to include in the final app - such as the data files and (hidden) imports mentioned above.
This repository is a collection of hooks for many packages, and allows PyInstaller to work with these packages seamlessly.
pyinstaller-hooks-contrib
is automatically installed when you install PyInstaller, or can be installed with pip:
pip install -U pyinstaller-hooks-contrib
Either a-package
works fine without a hook, or no-one has contributed hooks.
If you'd like to add a hook, or view information about hooks,
please see below.
If you've got a hook you want to share then great! The rest of this page will walk you through the process of contributing a hook. If you've been here before then you may want to skip to the summary checklist
Unless you are very comfortable with git rebase -i
, please provide one hook per pull request!
If you have more than one then submit them in separate pull requests.
Fork this repo if you haven't already done so.
(If you have a fork already but its old, click the Fetch upstream button on your fork's homepage.)
Clone and cd
inside your fork by running the following (replacing bob-the-barnacle
with your github username):
git clone https://github.com/bob-the-barnacle/pyinstaller-hoooks-contrib.git
cd pyinstaller-hooks-contrib
Create a new branch for you changes (replacing foo
with the name of the package):
You can name this branch whatever you like.
git checkout -b hook-for-foo
If you wish to create a virtual environment then do it now before proceeding to the next step.
Install this repo in editable mode.
This will overwrite your current installation.
(Note that you can reverse this with pip install --force-reinstall pyinstaller-hooks-contrib
).
pip install -e .
pip install -r requirements-test.txt
pip install flake8
Note that on macOS and Linux, pip
may by called pip3
.
If you normally use pip3
and python3
then use pip3
here too.
You may skip the 2nd line if you have no intention of providing tests (but please do provide tests!).
Standard hooks live in the src/_pyinstaller_hooks_contrib/hooks/stdhooks/ directory. Runtime hooks live in the src/_pyinstaller_hooks_contrib/hooks/rthooks/ directory. Simply copy your hook into there. If you're unsure if your hook is a runtime hook then it almost certainly is a standard hook.
Please annotate (with comments) anything unusual in the hook. Unusual here is defined as any of the following:
- Long lists of
hiddenimport
submodules. If you need lots of hidden imports then usecollect_submodules('foo')
. For bonus points, track down why so many submodules are hidden. Typical causes are:- Lazily loaded submodules (
importlib.importmodule()
inside a module__getattr__()
). - Dynamically loaded backends.
- Usage of
Cython
or Python extension modules containingimport
statements.
- Lazily loaded submodules (
- Use of
collect_all()
. This function's performance is abismal and it is broken by design because it confuses packages with distributions. Check that you really do need to collect all of submodules, data files, binaries, metadata and dependencies. If you do then add a comment to say so (and if you know it - why). Do not simply usecollect_all()
just to future proof the hook. - Any complicated
os.path
arithmetic (by which I simply mean overly complex filename manipulations).
All source files must contain the copyright header to be covered by our terms and conditions.
If you are adding a new hook (or any new python file), copy/paste the appropriate copyright header (below) at the top replacing 2021 with the current year.
GPL 2 header for standard hooks or other Python files.
# ------------------------------------------------------------------
# Copyright (c) 2021 PyInstaller Development Team.
#
# This file is distributed under the terms of the GNU General Public
# License (version 2.0 or later).
#
# The full license is available in LICENSE.GPL.txt, distributed with
# this software.
#
# SPDX-License-Identifier: GPL-2.0-or-later
# ------------------------------------------------------------------
APL header for runtime hooks only. Again, if you're unsure if your hook is a runtime hook then it'll be a standard hook.
# ------------------------------------------------------------------
# Copyright (c) 2021 PyInstaller Development Team.
#
# This file is distributed under the terms of the Apache License 2.0
#
# The full license is available in LICENSE.APL.txt, distributed with
# this software.
#
# SPDX-License-Identifier: Apache-2.0
# ------------------------------------------------------------------
If you are updating a hook, skip this step. Do not update the year of the copyright header - even if it's out of date.
Having tests is key to our continuous integration. With them we can automatically verify that your hook works on all platforms, all Python versions and new versions of libraries as and when they are released. Without them, we have no idea if the hook is broken until someone finds out the hard way. Please write tests!!!
Some user interface libraries may be impossible to test without user interaction or a wrapper library for some web API may require credentials (and possibly a paid subscription) to test. In such cases, don't provide a test. Instead explain either in the commit message or when you open your pull request why an automatic test is impractical then skip on to the next step.
A test should be the least amount of code required to cause a breakage
if you do not have the hook which you are contributing.
For example if you are writing a hook for a library called foo
which crashes immediately under PyInstaller on import foo
then import foo
is your test.
If import foo
works even without the hook then you will have to get a bit more creative.
Good sources of such minimal tests are introductory examples
from the documentation of whichever library you're writing a hook for.
Package's internal data files and hidden dependencies are prone to moving around so
tests should not explicitly check for presence of data files or hidden modules directly -
rather they should use parts of the library which are expected to use said data files or hidden modules.
Tests currently all live in src/_pyinstaller_hooks_contrib/tests/test_libraries.py.
Navigate there and add something like the following, replacing all occurrences of foo
with the real name of the library.
(Note where you put it in that file doesn't matter.)
@importorskip('foo')
def test_foo(pyi_builder):
pyi_builder.test_source("""
# Your test here!
import foo
foo.something_fooey()
""")
If the library has changed significantly over past versions then you may need to add version constraints to the test.
To do that, replace the @importorskip("foo")
with a call to PyInstaller.utils.tests.requires()
(e.g.
@requires("foo >= 1.4")
) to only run the test if the given version constraint is satisfied.
Note that @importorskip
uses module names (something you'd import
) whereas @requires
uses distribution names
(something you'd pip install
) so you'd use @importorskip("PIL")
but @requires("pillow")
.
For most packages, the distribution and packages names are the same.
Running our full test suite is not recommended as it will spend a very long time testing code which you have not touched.
Instead, run tests individually using either the -k
option to search for test names:
pytest -k test_foo
Or using full paths:
pytest src/_pyinstaller_hooks_contrib/tests/test_libraries.py::test_foo
Get the version of the package you are working with (pip show foo
)
and add it to the requirements-test-libraries.txt file.
The requirements already in there should guide you on the syntax.
To test hooks on all platforms we use Github's continuous integration (CI/CD). Our CI/CD is a bit unusual in that it's triggered manually and takes arguments which limit which tests are run. This is for the same reason we filter tests when running locally - the full test suite takes ages.
First push the changes you've made so far.
git push --set-upstream origin hook-for-foo
Replace billy-the-buffalo with your Github username in the following url then open it.
It should take you to the oneshot-test
actions workflow on your fork.
You may be asked if you want to enable actions on your fork - say yes.
https://github.com/billy-the-buffalo/pyinstaller-hooks-contrib/actions/workflows/oneshot-test.yml
Find the Run workflow button and click on it. If you can't see the button, select the Oneshot test tab from the list of workflows on the left of the page and it should appear. A dialog should appear containing one drop-down menu and 5 line-edit fields. This dialog is where you specify what to test and which platforms and Python versions to test on. Its fields are as follows:
- A branch to run from. Set this to the branch which you are using (e.g.
hook-for-foo
), - Which package(s) to install and their version(s).
Which packages to test are inferred from which packages are installed.
You can generally just copy your own changes to the
requirements-test-libraries.txt
file into this box.- Set to
foo
to test the latest version offoo
, - Set to
foo==1.2, foo==2.3
(note the comma) to test two different versions offoo
in separate jobs, - Set to
foo bar
(note the lack of a comma) to testfoo
andbar
in the same job,
- Set to
- Which OS or OSs to run on
- Set to
ubuntu
to test onlyubuntu
, - Set to
ubuntu, macos, windows
(order is unimportant) to test all three OSs.
- Set to
- Which Python version(s) to run on
- Set to
3.9
to test only Python 3.9, - Set to
3.6, 3.7, 3.8, 3.9
to test all currently supported version of Python.
- Set to
- The final two options can generally be left alone.
Hit the green Run workflow button at the bottom of the dialog, wait a few seconds then refresh the page. Your workflow run should appear.
We'll eventually want to see a build (or collection of builds) which pass on all OSs and all Python versions. Once you have one, hang onto its URL - you'll need it when you submit the pull request. If you can't get it to work - that's fine. Open a pull request as a draft, show us what you've got and we'll try and help.
If you find repeatedly entering the configuration into Github's Run workflow dialog arduous
then we also have a CLI script to launch it.
Run python scripts/cloud-test.py --help
which should walk you through it.
You will have to enter all the details again but, thanks to the wonders of terminal history,
rerunning a configuration is just a case of pressing up then enter.
We use flake8
to enforce code-style.
pip install flake8
if you haven't already then run it with the following.
Note that this assumes that you did create a new branch in the setup step.
git diff -U0 master | flake8 --diff -
No news is good news.
If it complains about your changes then do what it asks then run it again.
If you don't understand the errors it come up with them lookup the error code
in each line (a capital letter followed by a number e.g. W391
).
If it complains about code which you haven't written,
or if you didn't create a new branch at the start then, using git log
,
find the commit ID of the newest commit which you didn't write, copy it
and replace master
in the above command with that ID.
git diff -U0 a5d3841c282fa23fd68c3d6a85519e73c08acb4a | flake8 --diff -
Please do not fix flake8 issues found in parts of the repository other than the bit that you are working on. Not only is it very boring for you, but it is harder for maintainers to review your changes because so many of them are irrelevant to the hook you are adding or changing.
Please read news/README.txt before submitting you pull request. This will require you to know the pull request number before you make the pull request. You can usually guess it by adding 1 to the number of the latest issue or pull request. Alternatively, submit the pull request as a draft, then add, commit and push the news item after you know your pull request number.
A brief checklist for before submitting your pull request:
- All new Python files have the appropriate copyright header.
- You have written a news entry.
- Your changes satisfy the linter (run
git diff -U0 master | flake8 --diff -
). - You have written tests (if possible), pinned the test requirement and linked to a successful CI build.
Once you've done all the above, go ahead and create a pull request.
If you're stuck doing any of the above steps, create a draft pull request and explain what's wrong - we'll sort you out...
Feel free to copy/paste commit messages into the Github pull request title and description.
If you have run CI/CD, please include a link to it in your description so that we can see that it works.
If you've never done a pull request before, note that you can edit it simply by running git push
again.
No need to close the old one and start a new one.
If you plan to contribute frequently or are interested in becoming a developer,
send an email to [email protected]
to let us know.