Thanks for being here! Fern gives a lot of importance to being a community project, and we rely on your help as much as you rely on ours. If you have any feedback on what we could improve, please open an issue to discuss it!
All contributions start with an issue. Even if you have a good idea of what the problem is and what the solution looks like, please open an issue. This will give us an opportunity to align on the problem and solution, and to deconflict in the case that somebody else is already working on it.
Review our documentation! We appreciate any help we can get that makes our documentation more digestible.
Talk about Fern in your local meetups! Even our users aren't always aware of some of our features. Learn, then share your knowledge with your own circles.
Write code! We've got lots of open issues - feel free to volunteer for one by commenting on the issue.
Our documentation is powered by Fern's Docs product. All of the configuration for the docs lives in docs.yml.
To edit the docs, you can modify docs.yml
or any of the markdown that it references.
To validate that the docs, run:
npm install -g fern-api
fern check
When you make a PR to update the docs, a PR preview link will be generated which will allow you to test if your changes came out as intended.
Our repo is a monorepo that relies on Yarn workspaces and Yarn Plug'n'Play to run smoothly.
To get started:
Step 1: Fork this repo
Fork by clicking here.
Step 2: Clone your fork and open in VSCode
git clone <your fork>
cd fern
code .
Step 3: Install dependencies
yarn
Step 4: Use the "workspace" version of Typescript
- Open any TypeScript file in VSCode
- Open the Command Palette (Cmd+Shift+P on Mac) and select
Typescript: Select TypeScript Version...
- Choose
Use Workspace Version
This tells VSCode to rely on the version of TypeScript that lives in .yarn/sdks/typescript
, which is modified to work with Yarn PNP.
To compile the packages in this monorepo, run yarn compile
.
This repo contains both unit tests and integration (end-to-end) tests.
To run the unit tests: yarn test
.
To run the integration tests: yarn test:ete
.
Many of our tests rely on Jest snapshot testing. To rewrite snapshots, use -u
: yarn test -u
and yarn test:ete -u
.
To build the CLI, run either:
-
yarn dist:cli:dev
. This compiles and bundles a CLI that communicates with our dev cloud environment. The CLI is outputted topackages/cli/cli/dist/dev/cli.cjs
. -
yarn dist:cli:prod
. This compiles and bundles a CLI that communicates with our production cloud environment. The CLI is outputted topackages/cli/cli/dist/prod/cli.cjs
.
To run the locally-generated CLI, run:
FERN_NO_VERSION_REDIRECTION=true node <path to CLI> <args>
Fern generators read in IR (Intermediate Representation) and spit out
generated files. The IR is a JSON data structure that includes information
about your API and any additional information that may be convenient for a
code generator. For example, the IR includes all possible casings of every
string (e.g. snake_case
, camelCase
, PascalCase
) so that the
generators don't need to implement this individually.
As we add more features to the API definition, we introduce new versions of the IR. For example, if we wanted to add a new auth mechanism, we would eventually need to add it to the IR so that the generators could generate relevant code.
Each generator is pinned to an IR Version. Different versions of the generator, can dependend on differnt versions of the IR. For example, the Python SDK generator released 2 months ago depends on an older IR than the one released this week.
Note: The IR schema is modeled as a Fern Definition and you can see several versions of them in the
./fern
folder.
The Fern CLI should be able to run old generators so whenver we introduce a new IR version, we write a migration. In other words if you introduce IR V20, then you will have to write a migration from IR V20 -> IR V19 so that any generator that depends on a lower IR version can continue to be run from our CLI.
Step 1: Define the new IR
- Create a new Fern Definition for the IR version
fern/ir-types-vXXX
. Copy the latest IR Fern Definition as a starting point. - Introduce any changes you want in the new IR Fern Definition.
- Generate a TypeScript SDK for the IR by running
fern generate --api ir-types-vXXX
- Update all
package.json
files to use newir-sdk
npm version. Runyarn install
- Run
yarn compile
. You will see compile errors related to your schema changes.
Step 2: Write a reverse migration
In the ir-migrations
package, introduce a new migration.
You can copy the latest migration as a starting point.
To test our generators we have built a CLI tool called seed.
Seed handles building the generators from source and running them against all of the test definitions that are present in the repository. It also handles running scripts against the generated code to make sure that all the generated code compiles and works as intended.
To build seed, simply run
yarn seed:build
Each generator has a folder in the top level seed
directory. For example, the folder
for the typescript sdk generator is seed/ts-sdk
. This folder contains a config file
called seed.yml
as well as all the generated code for each test case.
To trigger seed tests on a specific generator run
yarn seed test --generator python-sdk --fixture file-download --skip-scripts
You can specify as many fixtures as you want. If you don't specify one, it will run on all the fields available.
It may be valuable to run seed on a particular Fern definition or OpenAPI spec. To do this,
you can use the seed run
command and point it at the fern folder:
yarn seed run --generator ts-sdk --path /Users/jdoe/fern
If the fern folder that you are pointing to has multiple APIs, then you must point it at the specific API that you are looking to generate:
yarn seed run --generator ts-sdk --path /Users/jdoe/fern/apis/imdb
To run against a custom fixture with an audience, run
yarn seed run --generator ts-sdk --path /Users/jdoe/fern/apis/imdb --audience external
By default, seed will build the docker container for the generator and execute the docker. Building a docker
adds extra time to your iteration cycle so we also have a mode to run the generators directly from source. All you
have to do is use the --local
flag.
For example, to run the TypeScript SDK generator from source, you can:
yarn seed test --generator ts-sdk --fixture imdb --local
The local flag will only work if the generator has configured the local
configuration in its seed.yml.
See here.