Language-agnostic composite action that executes a project's test suite with optional setup and failure-artifact upload. Designed for Forgejo Actions (works on GitHub Actions too).
## Inputs
| Input | Required | Default | Description |
|---|---|---|---|
| `command` | yes | — | Test command to execute (multiline allowed) |
| `setup` | no | `""` | Setup commands run before tests (install deps, build, etc.) |
| `working-directory` | no | `.` | Directory to run setup and command in |
| `shell` | no | `bash` | Shell for setup and command |
| `artifacts-path` | no | `""` | Glob of paths to upload on failure. Empty disables upload. |
| `artifacts-name` | no | `test-artifacts` | Name of the uploaded artifact bundle |
## Behavior
1. Runs `setup` if provided. Non-zero exit fails the job immediately.
2. Runs `command`. Exit code is captured but not yet propagated.
3. If `command` failed and `artifacts-path` is set, uploads matching files.
4. Re-exits with failure if `command` failed.
This ordering ensures artifacts are always uploaded on failure, even though the job ultimately fails.
Each example below is a complete workflow — drop it into `.forgejo/workflows/test.yml` (or `.github/workflows/test.yml`) and adjust `runs-on` to match a label one of your runners is registered with. See [Choosing a runner](#choosing-a-runner) below.
`runs-on` must match the labels a runner was registered with — there is no fuzzy match. If your runners are registered as `docker-amd64` and `docker-arm64`, then `runs-on: docker` will queue forever waiting for a runner that doesn't exist.
### Single architecture
Pick the label that matches your runner:
```yaml
jobs:
test:
runs-on: docker-amd64
```
### Both architectures (matrix)
Run the same job on every arch in parallel. Useful when you ship binaries, link against native libraries, or want to catch arch-specific bugs early:
`fail-fast: false` lets the other arch finish even if one fails, so you see both results.
### Does the architecture matter for the tests?
- **Pure JVM, Python, or Node** with no native dependencies: only execution speed differs.
- **Native code** (Rust, C/C++, JNI, Python wheels like `numpy`/`cryptography`, Node modules built with `node-gyp`): the compiled artifact is arch-specific. Same source, different binary — running the matrix exercises both.
- **Container images pulled during setup**: must publish a manifest for the runner's arch. Most official images do; some niche ones are amd64-only and will fail on arm64.
-`runs-on` must match a label your runner was registered with — see [Choosing a runner](#choosing-a-runner). The GitHub-style `ubuntu-latest` typically does not exist on Forgejo.
- Action references must use full URLs on Forgejo (`https://code.forgejo.org/actions/checkout@v4`), unlike GitHub where `actions/checkout@v4` is shorthand.
- The failure-artifact upload uses [`forgejo/upload-artifact`](https://code.forgejo.org/forgejo/upload-artifact), which is API-compatible with `actions/upload-artifact` and works on both Forgejo and GitHub.