|
| 1 | +<div align="center"> |
| 2 | + |
| 3 | +<h1>GitHub Workflows</h1> |
| 4 | + |
| 5 | +My GitHub workflows. Centralized for reuse. |
| 6 | + |
| 7 | +</div> |
| 8 | + |
| 9 | +**Table of contents** |
| 10 | + |
| 11 | +- [Published workflows](#published-workflows) |
| 12 | + - [npm-ci-build](#npm-ci-build) |
| 13 | +- [Goal: Universal targets](#goal-universal-targets) |
| 14 | + - [Industry observations](#industry-observations) |
| 15 | + - [My universal targets](#my-universal-targets) |
| 16 | + - [My universal output locations](#my-universal-output-locations) |
| 17 | +- [Style conventions](#style-conventions) |
| 18 | + |
| 19 | +## Published workflows |
| 20 | + |
| 21 | +> [!NOTE] |
| 22 | +> |
| 23 | +> The workflows use file versioning (`name~v1.yaml`) instead of repository tags. |
| 24 | +> |
| 25 | +> <details><summary>Why?</summary><div> |
| 26 | +> |
| 27 | +> This provides independent versioning of the workflows while maintaining them in one repository. |
| 28 | +> It also allows supporting older versions on the <code>main</code> branch (for example, upgrading all checkout actions). |
| 29 | +> |
| 30 | +> File versioning uses the tilde (`~`) because GitHub prohibits `@` in workflow file names. |
| 31 | +> (GitHub uses `@` for repository references.) |
| 32 | +> The tilde still reads well, displays well in URLs, and is not often used in file names. |
| 33 | +> |
| 34 | +> </div></details> |
| 35 | +
|
| 36 | +### npm-ci-build |
| 37 | + |
| 38 | +Job that executes `ci-build` logic for npm packages. |
| 39 | + |
| 40 | +```yaml |
| 41 | +jobs: |
| 42 | + ci-build: |
| 43 | + uses: connorjs/github-workflows/.github/workflows/npm-ci-build~v1.yaml |
| 44 | +``` |
| 45 | +
|
| 46 | +`v1` runs `ci-build` directly and assumes that the underlying package orchestrates the full build correctly. |
| 47 | + |
| 48 | +<details> |
| 49 | +<summary>Show inputs, outputs, assumptions, and notes</summary> |
| 50 | + |
| 51 | +#### Inputs |
| 52 | + |
| 53 | +| Name | Default | Type | Description | |
| 54 | +|:-------------:|:---------:|:-------:|:-------------------------------------------------------------------------------------------:| |
| 55 | +| `hasTests` | `true` | boolean | Whether the project has tests.<br>Used to skip test reporting for projects with no tests. | |
| 56 | +| `projectType` | `library` | choice | The type of project: `application` or `library`.<br>Used to customize the CI build process. | |
| 57 | + |
| 58 | +#### Outputs |
| 59 | + |
| 60 | +| Name | Description | |
| 61 | +|:--------:|:--------------------------------------------------------------------:| |
| 62 | +| `semVer` | The SemVer version number of this build (automated with GitVersion). | |
| 63 | + |
| 64 | +#### Assumptions |
| 65 | + |
| 66 | +The job makes the following assumptions about the repository. |
| 67 | + |
| 68 | +- Uses a `.node-version` file in the root of the repository to set the Node.js version. |
| 69 | +- Uses npm (not yarn or pnpm). |
| 70 | +- The `package.json` uses `0.0.0-gitversion` for its version (enables GitVersion automation). |
| 71 | + |
| 72 | +#### Notes |
| 73 | + |
| 74 | +The job includes automatic versioning via GitVersion. |
| 75 | +It will replace `0.0.0-gitversion` with the correct version. |
| 76 | +_Note: While another tool could replace GitVersion, the automatic version string will remain `0.0.0-gitversion`._ |
| 77 | + |
| 78 | +Note: The job will ignore any local `GitVersion.yaml`; it configures GitVersion for continuous deployment internally. |
| 79 | +Use `+semver:(major|minor)` in commit messages appropriately. |
| 80 | +(Patch happens automatically.) |
| 81 | + |
| 82 | +</details> |
| 83 | + |
| 84 | +## Goal: Universal targets |
| 85 | + |
| 86 | +I find that a universal set of targets (also called tasks) simplifies polyglot development. |
| 87 | +I observed the benefits with Amazon’s internal build system, and I built similar mechanisms at Total Quality Logistics (TQL). |
| 88 | + |
| 89 | +This repository represents my personal take on the idea, implemented via a consistent set of GitHub workflows. |
| 90 | + |
| 91 | +### Industry observations |
| 92 | + |
| 93 | +This section details my observations from industry tools that I have used. |
| 94 | + |
| 95 | +- Amazon: Standardized install, build, test (unit-test and integration-test) all under release. |
| 96 | + (Source: per my memory 18+ months later) |
| 97 | + |
| 98 | +- Gradle: Standardizes check, assemble, and build tasks. |
| 99 | + Includes some sort of standardized dependency installation (but I cannot find the task name). |
| 100 | + Specific plugins (such as java) extend with their own standard set (compile, javadoc, and more). |
| 101 | + (Source: [base plugin](https://docs.gradle.org/current/userguide/base_plugin.html)) |
| 102 | + |
| 103 | +- Maven: Standardizes validate, compile, test, package, verify, install, and deploy |
| 104 | + (Source: [Maven build lifecycle](https://maven.apache.org/guides/introduction/introduction-to-the-lifecycle.html)) |
| 105 | + |
| 106 | +- .NET (`dotnet`): Restore, build, test, format, run, and publish. |
| 107 | + (Source: My limited usage and [dotnet commands](https://learn.microsoft.com/en-us/dotnet/core/tools/dotnet)) |
| 108 | + |
| 109 | +- npm: Standardizes install/ci/rebuild, start/stop/restart, test, and many related to prepare and publish. |
| 110 | + Many libraries suggest build and lint, but there are others related (compile, tsc, specific linters, and formatters). |
| 111 | + (Source: My experience and [npm lifecycle scripts](https://docs.npmjs.com/cli/v10/using-npm/scripts)) |
| 112 | + |
| 113 | +- Python: Unsure. |
| 114 | + I know of install from pipenv. |
| 115 | + I see build, fmt, and test from hatch, install and build from flit, and easy_install and build from Setuptools. |
| 116 | + |
| 117 | +Each of these also had the concept of clean. |
| 118 | + |
| 119 | +### My universal targets |
| 120 | + |
| 121 | +Combining my experiences predominantly with Java, npm, and .NET, I arrive at the following universal targets. |
| 122 | +Each underlying build system should be able to hook into these standardized targets. |
| 123 | + |
| 124 | +- `install`: Installs (or restores) dependencies. |
| 125 | + |
| 126 | +- `build`: Builds the project. |
| 127 | + |
| 128 | +- `format`: Formats and lints the project. |
| 129 | + |
| 130 | +- `test`: Runs all tests that do not require a network connection. |
| 131 | + Often thought of as unit tests, but can contain network-free integration tests (aka component or module tests). |
| 132 | + This should include code coverage. |
| 133 | + |
| 134 | +- `publish`: Publish the library to its package manager. |
| 135 | + |
| 136 | +- `version`: Handles automatic versioning, if desired. |
| 137 | + |
| 138 | +- `clean`: Cleans any build or test artifacts. |
| 139 | + SHOULD NOT remove dependencies. |
| 140 | + |
| 141 | +In CI (think code reviews or pull requests), the “build” executes many of these targets. |
| 142 | +Using the term “build” here would overload the target name though. |
| 143 | +Therefore, I use the term `ci-build` to group all targets that should run during the “CI Build.” |
| 144 | + |
| 145 | +We called this `release` at Amazon, but I have found that folks think `release` means “make the release; actually publish,” which represents the wrong conclusion. |
| 146 | +Hence, I chose the term `ci-build`. |
| 147 | +_(I also considered terms such as `full-build`, `release-build`, and `ci` or replacing the underlying `build` term with `compile`. |
| 148 | +However, others came to the right conclusion with `ci-build`, which reinforces its choice.)_ |
| 149 | + |
| 150 | +If I was building my own CLI that wrapped underlying tools with these universal targets, I would include `compile` and `lint` as aliases for `build` and `format`. |
| 151 | +If both targets were present, the main name would take precedence. |
| 152 | + |
| 153 | +### My universal output locations |
| 154 | + |
| 155 | +Standardizing the output locations also simplifies polyglot development. |
| 156 | +While tooling (GitHub workflows) receives the most benefits, developers can benefit from standardized code coverage output locations. |
| 157 | +Standardized outputs can also simplify `.gitignore` configuration. |
| 158 | +_(Nit-pick side note: The 100s-lines-long default `.gitignore` files pain me.)_ |
| 159 | + |
| 160 | +The standard output locations follow. |
| 161 | + |
| 162 | +- `artifacts` should contain as much output as possible. |
| 163 | + This also simplifies clean. |
| 164 | + |
| 165 | +- Dependencies will use underlying tool conventions (example: `node_modules` for npm). |
| 166 | + |
| 167 | +- Publish will use underlying tool conventions where appropriate. |
| 168 | + Use `artifacts/dist` to vend transpiled sources. |
| 169 | + |
| 170 | +- Mono-repos will use `$projectName` in the path per artifact type (example: `artifacts/dist/$projectName`). |
| 171 | + This preserves output structure semantics between polyrepo (single project repositories) and monorepo use cases. |
| 172 | + |
| 173 | +- Tests will use `artifacts/test-results`. |
| 174 | + Prefer JUnit format for the test results file and use the name `test-result.xml`. |
| 175 | + Prefer Cobertura format for the code coverage results file and use the name `cobertura.xml`. |
| 176 | + |
| 177 | + In monorepos, each project likely has these files in `artifacts/test-results/$projectName` _AND_ the monorepo workspace aggregate in `artifacts/test-results` (N+1 output files). |
| 178 | + Note that tools like [reportgenerator](https://reportgenerator.io) exist to simplify aggregation. |
| 179 | + |
| 180 | +- The HTML output from code coverage should have the name `artifacts/coverage/index.html`. |
| 181 | + This differs from `artifacts/test-results` to simplify developers opening the report and to prevent conflicts in monorepos. |
| 182 | + |
| 183 | +## Style conventions |
| 184 | + |
| 185 | +Some notes on how I style (format) my GitHub workflow files. |
| 186 | + |
| 187 | +- The following naming rules ensure seamless script usage, such as in `actions/github-script`. |
| 188 | + - Prefer `PascalCase` for job and step IDs. |
| 189 | + - Prefer `camelCase` for variable names (inputs and outputs). |
| 190 | + - _This means avoid `snake_case`, `kebab-case`, `TRAIN-CASE`, or `SCREAM_CASE` unless a 3rd party uses them._ |
| 191 | + |
| 192 | +- Format expressions with inner spaces: `${{ foo.bar }}`. |
| 193 | + |
| 194 | +- Prefer no quotes if YAML does not require them. |
| 195 | + |
| 196 | +- Automatically format files: Use prettier, an editor using .editorconfig, or a similar formatter. |
| 197 | + |
| 198 | +- Start each step with `name`. |
| 199 | + - Use sentence case for `name` (with allowed use of Proper Nouns when it helps such as GitVersion). |
| 200 | + - Exception: Prefer “CI Build” (instead of “CI build” or “ci build”). |
| 201 | + |
| 202 | +- If a step (or job) would benefit from a longer description, include a YAML comment on the line after `name` (unless GitHub adds proper `description` field). |
0 commit comments