tree: 94317b8aef56f37d9eda85dd1fd89ac30d08255a [path history] [tgz]
  1. config/
  2. .gitattributes
  3. .gitignore
  4. config.yml
  5. Makefile
  6. README.md
.circleci/README.md

How to use CircleCI multi-file config

This README and the Makefile should be in your .circleci directory, in the root of your repository. All path references in this README assume we are in this .circleci directory.

The Makefile in this directory generates ./config.yml in CircleCI 2.0 syntax, from the tree rooted at ./config/, which contains files in CircleCI 2.0 or 2.1 syntax.

Quickstart

The basic workflow is:

  • Edit source files in ./config/
  • When you are done, run make ci-config to update ./config.yml
  • Commit this entire .circleci directory, including that generated file together.
  • Run make ci-verify to ensure the current ./config.yml is up to date with the source.

When merging this .circleci directory:

  • Do not merge the generated ./config.yml file, instead:
  • Merge the source files under ./config/, and then
  • Run make ci-config to re-generate the merged ./config.yml

And that's it, for more detail, read on!

How does it work, roughly?

CircleCI supports generating a single config file from many, using the $ circleci config pack command. It also supports expanding 2.1 syntax to 2.0 syntax using the $ circleci config process command. We use these two commands, stitched together using the Makefile to implement the workflow.

Prerequisites

You will need the CircleCI CLI tool installed and working, at least version 0.1.5607. You can download this tool directly from GitHub Releases.

$ circleci version
0.1.5607+f705856

Updating the config source

Before making changes, be sure to understand the layout of the ./config/ file tree, as well as circleci 2.1 syntax. See the Syntax and layout section below.

To update the config, you should edit, add or remove files in the ./config/ directory, and then run make ci-config. If that's successful, you should then commit every *.yml file in the tree rooted in this directory. That is: you should commit both the source under ./config/ and the generated file ./config.yml at the same time, in the same commit. The included git pre-commit hook will help with this. Do not edit the ./config.yml file directly, as you will lose your changes next time make ci-config is run.

Verifying ./config.yml

To check whether or not the current ./config.yml is up to date with the source and valid, run $ make ci-verify. Note that $ make ci-verify should be run in CI, in case not everyone has the git pre-commit hook set up correctly.

Example shell session

$ make ci-config
config.yml updated 
$ git add -A . # The -A makes sure to include deletions/renames etc.
$ git commit -m "ci: blah blah blah"
Changes detected in .circleci/, running 'make -C .circleci ci-verify'
--> Generated config.yml is up to date!
--> Config file at config.yml is valid.

Syntax and layout

It is important to understand the layout of the config directory. Read the documentation on packing a config for a full understanding of how multiple YAML files are merged by the circleci CLI tool.

Here is an example file tree (with comments added afterwards):

$ tree . 
.
├── Makefile
├── README.md # This file.
├── config    # The source code for config.yml is rooted here.
│   ├── @config.yml # Files beginning with @ are treated specially by `circleci config pack`
│   ├── commands    # Subdirectories of config become top-level keys.
│   │   └── go_test.yml  # Filenames (minus .yml) become top-level keys under
│   │   └── go_build.yml # their parent (in this case "commands").
                       # The contents of go_test.yml therefore are placed at: .commands.go_test:
│   └── jobs             # jobs also becomes a top-level key under config...
│       ├── build.yml    # ...and likewise filenames become keys under their parent.
│       └── test.yml
└── config.yml # The generated file in 2.0 syntax.

About those @ files... Preceding a filename with @ indicates to $ circleci config pack that the contents of this YAML file should be at the top-level, rather than underneath a key named after their filename. This naming convention is unfortunate as it breaks autocompletion in bash, but there we go.