11 KiB

Raw Blame History

(new-packages)=

Creating a Pyodide package

It is recommended to look into how other similar packages are built in Pyodide. If you encounter difficulties in building your package after trying the steps listed here, open a new Pyodide issue.

Quickstart

If you wish to use a package in Pyodide that is not already included, first you need to determine whether it is necessary to package it for Pyodide. Ideally, you should start this process with package dependencies.

1. Determining if creating a Pyodide package is necessary

Most pure Python packages can be installed directly from PyPI with {func}micropip.install if they have a pure Python wheel. Check if this is the case by going to the pypi.org/project/<package-name> URL and checking if the "Download files" tab contains a file that ends with *py3-none-any.whl.

If the wheel is not on PyPI, but nevertheless you believe there is nothing preventing it (it is a Python package without C extensions):

you can create the wheel yourself by running,
```
python -m pip install build
python -m build
```
from within the package folder where the setup.py are located. See the Python packaging guide for more details. Then upload the wheel file somewhere (not to PyPI) and install it with micropip via its URL.
you can also open an issue in the package repository asking the authors to upload the wheel.

If however the package has C extensions or its code requires patching, then continue to the next steps.

To determine if a package has C extensions, check if its `setup.py` contains
any compilation commands.

2. Creating the `meta.yaml` file

If your package is on PyPI, the easiest place to start is with the {ref}mkpkg tool <pyodide-mkpkg>. From the Pyodide root directory, install the tool with pip install -e pyodide-build, then run:

pyodide-build mkpkg <package-name>

This will generate a meta.yaml under packages/<package-name>/ (see {ref}meta-yaml-spec) that should work out of the box for many simple Python packages. This tool will populate the latest version, download link and sha256 hash by querying PyPI. It doesn't currently handle package dependencies, so you will need to specify those yourself.

You can also use the meta.yaml of other Pyodide packages in the packages/ folder as a starting point.

To reliably determine build and runtime dependencies, including for non Python
libraries, it is often useful to verify if the package was already built on
[conda-forge](https://conda-forge.org/) and open the corresponding `meta.yaml`
file. This can be done either by checking if the URL
`https://github.com/conda-forge/<package-name>-feedstock/blob/master/recipe/meta.yaml`
exists, or by searching the [conda-forge GitHub
org](https://github.com/conda-forge/) for the package name.

The `meta.yaml` in Pyodide was inspired by the one in conda, however it is
not strictly compatible.

3. Building the package and investigating issues

Once the meta.yaml is ready, build the package with the following commands from inside the package directory packages/<package-name>

export PYTHONPATH="$PYTHONPATH:/path/to/pyodide/pyodide-build/"
python -m pyodide_build buildpkg meta.yaml
cp build/*.data build/*.js ../../build/

and see if there are any errors.

If there are errors you might need to

patch the package by adding .patch files to packages/<package-name>/patches
add the patch files to the source/patches field in the meta.yaml file

then restart the build.

Generating patches

If the package has a git repository, the easiest way to make a patch is usually:

Clone the git repository. You might want to use the options git clone --depth 1 --branch <version-tag>. Find the appropriate tag given the version of the package you are trying to modify.
Make whatever changes you want. Commit them.
Use git format-patch HEAD~1 -o <pyodide-root>/packages/<package-name>/patches/ to generate a patch file for your changes and store it directly into the patches folder.

It is also possible to create a patch with git diff --no-index old_file new_file > my_changes.patch, but you will probably have to fix up the paths in the generated diff. If it is taking a large number of attempts, you can use the flags --src-prefix and --dst-prefix to try to ensure the patch file is generated with the correct paths.

The package build pipeline

Pyodide includes a toolchain to make it easier to add new third-party Python libraries to the build. We automate the following steps:

Download a source tarball (usually from PyPI)
Confirm integrity of the package by comparing it to a checksum
Apply patches, if any, to the source distribution
Add extra files, if any, to the source distribution
If the package includes C/C++/Cython extensions:
- Build the package natively, keeping track of invocations of the native compiler and linker
- Rebuild the package using emscripten to target WebAssembly
If the package is pure Python:
- Run the setup.py script to get the built package
Unvendor unit tests included in the installation folder to a separate package <package name>-tests
Package the results into an emscripten virtual filesystem package, which comprises:
- A .data file containing the file contents of the whole package, concatenated together
- A .js file which contains metadata about the files and installs them into the virtual filesystem.

Lastly, a packages.json file is output containing the dependency tree of all packages, so {any}pyodide.loadPackage can load a package's dependencies automatically.

Partial Rebuilds

By default, each time you run buildpkg, pyodide-build will delete the entire source directory and replace it with a fresh copy from the download url. This is to ensure build repeatability. For debugging purposes, this is likely to be undesirable. If you want to try out a modified source tree, you can pass the flag --continue and buildpkg will try to build from the existing source tree. This can cause various issues, but if it works it is much more convenient.

Using the --continue flag, you can modify the sources in tree to fix the build, then when it works, copy the modified sources into your checked out copy of the package source repository and use git format-patch to generate the patch.

For very complicated builds, we also have a mechanism for repeating from the replay stage of the build. If the build is failing during the replay stage, you will see lines like [line 766 of 1156] labeling which command is being replayed. --continue=replay will start over from the beginning of the replay stage, --continue=replay:766 will start from step 766 of the replay stage.

C library dependencies

Some Python packages depend on certain C libraries, e.g. lxml depends on libxml.

To package a C library, create a directory in packages/ for the C library. In the directory, you should write meta.yaml that specifies metadata about the library. See {ref}meta-yaml-spec for more details.

The minimal example of meta.yaml for a C library is:

package:
  name: <name>
  version: <version>

source:
  url: <url>
  sha256: <sha256>

requirements:
  run:
    - <requirement>

build:
  library: true
  script: |
    emconfigure ./configure
    emmake make -j ${PYODIDE_JOBS:-3}

You can use the meta.yaml of other C libraries such as libxml as a starting point.

After packaging a C library, it can be added as a dependency of a Python package like a normal dependency. See lxml and libxml for an example (and also scipy and CLAPACK).

Remark: Certain C libraries come as emscripten ports, and do not have to be built manually. They can be used by adding e.g. -s USE_ZLIB in the cflags of the Python package. See e.g. matplotlib for an example. The full list of libraries with Emscripten ports is here.

Structure of a Pyodide package

This section describes the structure of a pure Python package, and how our build system creates it. In general, it is not recommended to construct these by hand; instead create a Python wheel and install it with micropip.

Pyodide is obtained by compiling CPython into WebAssembly. As such, it loads packages the same way as CPython --- it looks for relevant files .py files in /lib/python3.x/. When creating and loading a package, our job is to put our .py files in the right location in emscripten's virtual filesystem.

Suppose you have a Python library that consists of a single directory /PATH/TO/LIB/ whose contents would go into /lib/python3.9/site-packages/PACKAGE_NAME/ under a normal Python installation.

The simplest version of the corresponding Pyodide package contains two files --- PACKAGE_NAME.data and PACKAGE_NAME.js. The first file PACKAGE_NAME.data is a concatenation of all contents of /PATH/TO/LIB. When loading the package via pyodide.loadPackage, Pyodide will load and run PACKAGE_NAME.js. The script then fetches PACKAGE_NAME.data and extracts the contents to emscripten's virtual filesystem. Afterwards, since the files are now in /lib/python3.9/, running import PACKAGE_NAME in Python will successfully import the module as usual.

To construct this bundle, we use the file_packager.py script from emscripten. We invoke it as follows:

$ ./tools/file_packager.sh \
     PACKAGE_NAME.data \
     --js-output=PACKAGE_NAME.js \
     --preload /PATH/TO/LIB/@/lib/python3.9/site-packages/PACKAGE_NAME/

The arguments can be explained as follows:

PACKAGE_NAME.data indicates where to put the data file
--js-output=PACKAGE_NAME.js indicates where to put the javascript file
--preload instructs the package to look for the file/directory before the separator @ (namely /PATH/TO/LIB/) and place it at the path after the @ in the virtual filesystem (namely /lib/python3.9/site-packages/PACKAGE_NAME/).

file_packager.sh adds the following options:

--lz4 to use LZ4 to compress the files
--export-name=globalThis.__pyodide_module tells file_packager where to find the main Emscripten module for linking.
--exclude *__pycache__* to omit the pycache directories
--use-preload-plugins says to automatically decode files based on their extension The preload plugins are important if the package contains any .so files because WebAssembly.instantiate is asynchronous.

.. toctree::
   :hidden:

   meta-yaml.md

11 KiB Raw Blame History