doc: add DEVELOPING.md with release workflow (#602)

Describe the release workflow as flowing from tag -> builds -> release.
This commit is contained in:
Chris Dickinson
2023-11-21 13:30:31 -08:00
committed by GitHub
parent 15c30dfa8c
commit 45b0749fe2

117
DEVELOPING.md Normal file
View File

@@ -0,0 +1,117 @@
# HACKING
## cutting releases
### goals
Cutting a release should be a boring, rote process with as little excitement as
possible. Following the processes in this document, we should be able to cut a
release at any time without worrying about producing bad artifacts. Our process
should let us resolve build issues without affecting library users.
### branching
1. The `main` branch represents the next major version of the library.
2. Previous major versions should be tracked using `v0.x`, `v1.x`, `v2.x`, used
for backporting changes as necessary.
3. Libraries should generate a `latest` release using, e.g.,
`marvinpinto/action-automatic-releases` on changes to the `main` branch.
### tag and release process
1. Pick a target semver value. Prepend the semver value with `v`: `v1.2.3`.
Increment the minor version for additive changes and patch for bugfixes.
- For trickier changes, consider using release candidates: `rc0`, `rc1`, etc.
2. Create an empty git commit for the tag to point at: `git commit -m 'v1.2.3-rc1' --allow-empty`.
3. Create a new tag against that commit: `v1.2.3-rc1`.
4. Push the changes to the library: `git push origin main v1.2.3-rc1`.
- You can separate these steps: `git push origin main` followed by `git push origin v1.2.3-rc1`,
if you want to make absolutely sure the commit you're pushing builds correctly before tagging it.
5. Wait for the tag `build` workflow to complete.
- The `build` workflow should create a _draft_ release (using `softprops/action-gh-release` with `draft`
set to `true`) and upload built artifacts to the release.
6. Once the workflow is complete, do whatever testing is necessary using the artifacts.
- TODO: We can add automation to this step so that we test on downstream deps automatically: e.g., if we
build a new kernel, we _should_ be able to trigger tests in the `python-sdk` _using_ that new kernel.
7. Once we're confident the release is good, go to the releases page for the library and edit the draft release.
- If the release is a release candidate (`rc0..N`), make sure to mark the release as a "prerelease".
- Publish the draft release.
- This kicks off the publication workflow: taking the artifacts built during the `build` workflow and publishing
them to any necessary registry or repository.
- In extism, this publishes `extism-maturin` to PyPI as `extism-sys` and the dotnet packages to nuget.
- In `python-sdk`, this publishes `extism` to PyPI.
- In `js-sdk`, this publishes `@extism/extism` (and `extism`) to NPM.
> **Note**
> If you're at all worried about a release, use a private fork of the target library repo to test the release first (e.g., `extism/dev-extism`.)
#### CLI flow
For official releases:
```
$ git commit -m 'v9.9.9' --allow-empty
$ git tag v9.9.9
$ git push origin main v9.9.9
$ gh run watch
$ gh release edit v9.9.9 --tag v9.9.9 --title 'v9.9.9' --draft=false
$ gh run watch
```
For prereleases:
```
$ git commit -m 'v9.9.9' --allow-empty
$ git tag v9.9.9
$ git push origin main v9.9.9
$ gh run watch
$ gh release edit v9.9.9 --tag v9.9.9 --title 'v9.9.9' --draft=false --prerelease
$ gh run watch
```
### implementation
Libraries should:
- Provide a `ci` workflow, triggered on PR and `workflow_dispatch`.
- This workflow should exercise the tests, linting, and documentation generation of the library.
- Provide a `build` workflow, triggered on `v*` tags and merges to `main`
- This workflow should produce artifacts and attach them to a draft release (if operating on a tag) or a `latest` release (if operating on `main`.)
- Artifacts include: source tarballs, checksums, shared objects, and documentation.
- Provide a `release` workflow, triggered on github releases:
- This workflow should expect artifacts from the draft release to be available.
- Artifacts from the release should be published to their final destination as part of this workflow: tarballs to NPM, documentation to Cloudflare R2/Amazon S3/$yourFavoriteBucket.
### A rough list of libraries and downstreams
```mermaid
flowchart TD;
A["runtime"] --> B["libextism"];
B --> C["extism-maturin"];
B --> X["nuget-extism"];
C --> D["python-sdk"];
B --> E["ruby-sdk"];
A --> F["go-sdk"];
G["plugins"] --> B;
G --> D;
G --> E;
G --> F;
G --> H["js-sdk"];
F --> I["cli"];
G --> J["dotnet-sdk"];
X --> J;
G --> K["cpp-sdk"];
G --> L["zig-sdk"];
B --> L;
G --> M["haskell-sdk"];
B --> M;
G --> N["php-sdk"];
B --> N;
G --> O["elixir-sdk"];
B --> O;
G --> P["d-sdk"];
B --> P;
G --> Q["ocaml-sdk"];
B --> Q;
```