packaging - conda-based napari installers

Important: This repo is used by napari core maintainers.

Description

This repository stores the tools (CI pipelines and helper scripts) needed to create conda-based napari installers for Linux, macOS and Windows. It uses the constructor tool which allows constructing an installer for a collection of conda packages.

Documentation

You can find more information in the following resources:

• Napari docs about packaging
• Napari docs about creating installers using this repo and the constructor tool
• NAP-2: Distributing napari with conda-based packaging

Usage

Quickstart

This is how to build an installer for the latest napari release. We assume that napari/napari and napari/packaging will be cloned as siblings to napari/ and napari-packaging, respectively:

git clone https://github.com/napari/napari.git napari  # skip if cloned already
cd napari
git checkout <latest-tag>  # only needed for some resources like images, not the code itself
cd ..
git clone https://github.com/napari/packaging.git napari-packaging
cd napari-packaging
conda env create -n napari-packaging-installers --file environments/ci_installers_environment.yml
conda activate napari-packaging-installers
pip install -e ../napari --no-deps
CONSTRUCTOR_PYTHON_VERSION="3.11" python build_installers.py --location ../napari
# Installers will be generated under `_work/`.

For any other versions, it's probably easier to run a workflow_dispatch trigger on .github/workflows/make_bundle_conda.yml.

Locally built packages

napari installers are created from conda packages exclusively, using constructor. This means that we need to have conda packages for the napari version we want to distribute. We get those from conda channels: conda-forge, napari, napari/label/nightly, or our own local channel.

Our Github workflows rely on conda-forge feedstock scripts to have the closes possible setup, but for debugging installers we can take a few shortcuts. We only need to know which version we have cloned:

conda activate napari-packaging-installers
CONSTRUCTOR_USE_LOCAL=1 python build_installers.py --version
# e.g. 0.6.5dev18+gc9111b7bb

Now, let's edit the recipe.yaml file so it points to our cloned version:

 context:
   name: napari
-  version: "REPLACE_ME"
+  version: "0.6.5dev18+gc9111b7bb"
   python_min: "3.10"
   build_number: 0

 recipe:
   name: ${{ name|lower }}
   version: ${{ version }}

 source:
-  # This section has been patched from the original conda-forge feedstock
-  # so it uses the mounted volume of the cloned repo in a Docker image
-  - path: /home/conda/feedstock_root/napari-source
+  - path: ../../napari

Now build the recipe, making sure packages are generated in _work/packages/:

conda create -n rattler-build rattler-build
conda activate rattler-build
CONDA_BLD_PATH=_work/packages/ rattler-build build --recipe conda-recipe/

Tip: You may skip tests with --test=skip

The installers can now be built from the local packages with:

conda activate napari-packaging-installers
CONSTRUCTOR_USE_LOCAL=1 CONDA_BLD_PATH=_work/packages/ CONSTRUCTOR_PYTHON_VERSION="3.11" python build_installers.py --location ../napari

The artifacts will be available in _work/.