D2's long-term mission is to significantly reduce the amount of time and effort it takes to create and maintain high-quality diagrams for every software team. We started this because we love the idea of creating diagrams with text -- but it was clear the existing solutions were inadequete in their state and speed of execution for this idea to be widely usable.
We've tried our best to avoid the mistakes of the past and take inspiration from the most successful modern programming and configuration languages. D2 has built up each step of the text-to-diagram pipeline from scratch, rethinking each one from first principles, from the dead simple syntax, to the readable compiler, our own SVG renderer, etc.
D2 is committed to making something people want to use. That means contributions don't only have to be in the form of pull requests. Your bug reports, plugins, examples, discussions of new ideas, help a great deal.
If you'd like to get involved, we're also committed to helping you merge that first pull request. You should be able to freely pick up Issues tagged as "good first issue". Those will usually include notes on getting started, but if they're missing or insufficient, just leave a comment and we'll add more detail. If you need help getting started, please don't hesitate to pop into Discord -- as long as you want to help, we'll find the perfect task (complexity matches your appetite and programming experience, in an area you're interested in, etc).
Most of D2's CI is open sourced in its own repository, included as a submodule. After you clone D2, make sure you initialize the submodules:
git submodule update --init --recursive./make.sh runs everything. Subcommands to run individual parts of the CI:
./make.sh fmt./make.sh lint./make.sh test./make.sh race./make.sh build
Here's what a successful run should look like:
Please make sure CI is passing for any PRs submitted for review.
Be sure to update the submodule whenever there are changes:
git submodule update --recursive- Use Go 1.22.
- Please sign your commits (#557 (comment)).
- D2 uses Issues as TODOs. No auto-closing on staleness.
- Branch off
master. - If there's an Issue related, include it by adding "[one-word] #[issue]", e.g. "Fixes #123" somewhere in the description.
- Whenever possible and relevant, include a screenshot or screen-recording.
Unless you've contributed before, it's safest to choose an Issue with a "good first issue" label. If you'd like to propose new functionality or change to current functionality, please create an Issue first with a proposal. When you start work on an Issue, please leave a comment so others know that it's being worked on.
D2 has extensive tests, and all code changes must include tests.
With the exception of changes to the renderer, all code should include a package-specific test. If it's a visual change, an end-to-end (e2e) test should accompany.
Let's say I make some code changes. I can easily see how this affects the end result by running:
./ci/e2ereport.sh -delta
This gives me a nice HTML output of what the test expected vs what it got (this was a PR fixing multi-byte character labels):
Run ./ci/e2ereport.sh -help for flags, including how to get deltas for a single test.
If you're testing labels and strings, it's encouraged to use 1-letter strings (x) in small
functional tests to minimally pinpoint issues. If you are testing something that exercises
variations in strings, or want to mimic more realistic diagram text, it's encouraged you
generate random strings and words from fortune. It gives a good range of the English
language. (Sometimes it gives controversial sentences -- don't use those.)
Script to generate one line of random text:
ipsum1() {
fortune | head -n1 | sed 's/^ *//;s/ *$//' | tr -d '\n' | tee /dev/stderr | pbcopy
}
Run: ./ci/test.sh
CI runs tests with -race to catch potential race conditions. It's much slower, but if
your machine can run it locally, you can do so with ./make.sh race.
If you add a new test and run, it will show failure. That's because the vast majority of
D2's tests are comparing outputs. You don't define the expected output manually. The
testing library generates it and it's checked into version control if it looks right. So
for the first run of a new test, it has no expected output, and will fail. To accept the
result as the expected, run the test with environment variable TESTDATA_ACCEPT=1.
D2 has chaos tests which produce random configurations of diagrams. It can be helpful to run a few iterations (N~=100) to spot cover your manual tests.
D2_CHAOS_MAXI=100 D2_CHAOS_N=100 ./ci/test.sh ./d2chaos
The code itself should be documented as appropriate with Go-style comments. No rules here,
GetX() doesn't need a // GetX gets X.
If it's some new functionality, please submit a pull request to document it in the language docs: https://github.com/terrastruct/d2-docs.
If you have any questions or would like to get more involved, feel free to open an issue to discuss publicly, or chat in Discord! If you have a private inquiry, feel free to email us at [email protected].