These are fully automated in GitHub Actions. Check the most recent run succeeded, and make a note of its commit, because we'll be tagging it later.
Download the maven artifact from GitHub Actions, and unpack it into the local maven
directory on all machines that will be used by the subsequent tests.
Open the demo app in Android Studio, and run the Java and Python unit tests on a representative device under the following conditions:
- Set build variant to "debug".
- Clean install, then run tests twice in the same process.
- Kill the process, then run tests again in a new process.
- Test "release" variant".
- Test "releaseMinify" variant (minify is not enabled in the "release" variant, because it could prevent users importing classes in the Python console).
- Change back to "debug" variant, and test a single-ABI build by temporarily changing abiFilters.
Download the demo artifact from GitHub Actions, and unpack the APK from it.
Install the APK and run the Java and Python unit tests on all ABIs, with at least one device being each of the following:
- A physical device (on all ABIs if possible)
- minSdk (on all ABIs if possible)
- targetSdk
- A clean install
- An upgrade from the previous public release, with the tests already run
Test all the UI elements of the app on both minSdkVersion and targetSdkVersion.
For each of the non-default Python versions, run the Java and Python unit tests on the same devices.
Open the pkgtest app in Android Studio, and temporarily edit the top-level build.gradle file to use the local Chaquopy version.
Record performance data in performance.md, and investigate if significantly worse than the previous version. Remember that the tests and the packages themselves may have changed.
Open the pkgtest app in Android Studio, and temporarily edit the top-level build.gradle file to use the local Chaquopy version.
Temporarily edit the app/build.gradle file to set PACKAGES to the top 40 recipes,
ordered by number of PyPI downloads, excluding TensorFlow (#1209):
- Get the PyPI statistics as described in server/pypi/README-internal.md.
cd server/pypi/packagescat pypi-downloads.csv | cut -d, -f1 | while read name; do if [ -e $name ] && [ $name != tensorflow ]; then echo $name; fi; done | head -n40 | tr '\n' ' '
As of 2023-12, this is:
numpy cryptography cffi pandas aiohttp yarl greenlet frozenlist grpcio lxml psutil multidict pillow scipy bcrypt matplotlib pynacl scikit-learn kiwisolver regex ruamel-yaml-clib google-crc32c pycryptodomex contourpy pyzmq pycryptodome zope-interface h5py tokenizers torch shapely numba llvmlite xgboost scikit-image statsmodels sentencepiece opencv-python torchvision brotli
Search the package test scripts for the word "Android", and consider adding any packages which test Chaquopy (as opposed to the package itself) in a way that isn't covered by Chaquopy's own unit tests.
Set abiFilters to each of armeabi-v7a and arm64-v8a (this tests the single-ABI
case), and test on those ABIs, with at least one device being each of the following:
- A physical device (on all ABIs if possible)
- minSdk (on all ABIs if possible)
- targetSdk
- A clean install
Set abiFilters to "x86", "x86_64" (this tests the multi-ABI case), and test on those
ABIs, with at least one device being each of the following:
- minSdk (on all ABIs if possible)
Use release/bundle.sh to create bundle JARs for the following things, and release them to Maven Central:
com.chaquo.python.gradle.plugingradleruntime/*target(if necessary)
As a backup, also upload them to https://chaquo.com/maven-central.
Make sure all public repositories are clean (git clean -xdf).
Run release/release_public.sh OLD_VER NEW_VER, where OLD_VER is the label in this
repository from which the public repositories were last updated.
If the script reports any files which require manual merging (e.g. build.gradle), examine them and update all the public repositories as necessary.
- The public apps should use an Android Gradle plugin version which is at least one year old, because new versions of AGP are incompatible with old versions of Android Studio (https://android-developers.googleblog.com/2020/12/announcing-android-gradle-plugin.html).
- If updating the AGP version, also update to the corresponding versions of these things
(see the integration tests):
- Gradle
- gradle.properties
- Kotlin plugin
- If .gitignore has changed, git rm any newly-ignored files.
Open each public app in Android Studio and test it on minSdk and targetSdk, with a clean install.
Close the projects to make sure .idea files are written.
Take the demo app APK which was tested above, and release it on Google Play, updating the description and screenshots if necessary.
Set reminder to check for Google Play crash reports regularly over the next month.
Add news fragments as necessary, and check the release notes by running towncrier build --version VERSION --draft. Once happy, save the fragments in a separate commit, because
they'll be deleted when running towncrier in non-draft mode.
Update:
changelog.rst, by running the abovetowncriercommand without--draft.versions.rstreleaseinconf.py
Run gradlew runtime:doc, and upload to server.
If major.minor version number has changed:
- Update "current" symlink (
ln -sfT). - Add link on WordPress documentation page.
Commit and push all example app repositories.
Tag the commit the GitHub Actions artifacts were built from, and push the tag.
Increment VERSION.txt for next version number.
Commit and push this repository.
Create release page on GitHub with a link to the change log section.
Post blog entry to website.
Post link to X, and enable "only accounts you mention can reply", because it doesn't reliably send email notifications of replies. Enabling replies and apparently ignoring them would make us look worse than not enabling replies at all.
Update any affected GitHub issues, StackOverflow questions, email threads, etc.
If there are any packages whose announcement was postponed until this release, go through the package release procedure in pypi/README-internal.md.