- Install Node 10 LTS and yarn. Older or newer versions may or may not work. (Recommend
nvm
andbrew install yarn --without-node
on mac.) - Install Docker.app. Our database and other services are configured to run in docker.
- Symlink
.env.example
to.env
, which sets up your environment to run from Docker. You can copy and modify.env.example
to.env
if the defaults won't work for you. - Start postgres with:
docker-compose up
. This will set up a postgres docker image and start it. - Run
yarn
to install dependencies. - Run
yarn db:create
to create development and test databases. - Run
yarn build
to build the application, including supporting scripts.
Note: Start docker-compose up
and leave it running any time you want to run the app/tests.
This repository includes a simple redux/graphql-based app. If you'd like to play around with this example, here's how to start it up:
- Run
yarn build
to, among other things, create generated types and scripts. - Run
yarn db:migrate:latest
to migrate your development and test databases. - Optionally run
yarn db:migrate-and-seed
to add some test data to your dev database. - Run
yarn dev
to start the hot-reloading dev server, which can be visited on port 3000. - See the interactive style guide and component tests by running
yarn dev:storybook
and visitlocalhost:9001
. - To run unit tests, run
yarn test:unit
oryarn test:unit --watch
for the interactive jest-based test runner.
For this project, we have decided not to check in the generated GraphQL types. This means that
you'll need to run yarn build
before you begin viewing or editing the source code, lest you see
spurious type errors and failed imports.
This project is a single-page webapp using the following technologies:
- TypeScript – a type-safe variant of JavaScript from Microsoft which reduces errors and improves IDE/editor support over regular JavaScript.
- Node.js – powers our server, and is pinned to the latest LTS release.
- Express – our HTTP server, which is lightly used only to host our GraphQL API.
- GraphQL – an alternative to REST apis which supports a demand-driven architecture. Our GraphQL server is Apollo GraphQL server.
- Jest for unit testing.
- Webpack – builds our application for our various deployment targets.
- Apollo Link State for client state management.
- Nightmare.js for acceptance testing.
- React Storybook for component documentation and style guides.
- JSVerify for property-based testing.
This repository is structured to encourage a view of the whole repository as one application. The client and server are “just” different entry points, and we use webpack to elide libraries and code that are irrelevant to a particular entry point.
There are a few key directories:
entry
– contains the primary entry points of the application. If you want to see what happens when you start up the client or server, start there. These are also the entry points for webpack.webpack
contains a webpack configuration for each entry point, as well aswebpack-dev-server.js
which sets up the dev server used during development.modules
contains all of the code. Each module is a self-contained concept that may be used by other modules, command-line scripts, etc.config
contains configuration files for our various environments. The default config is set up as a twelve-factor app to be hosted in heroku. Most variables can be controlled via the environment – seeconfig/default.js
.dist
is where webpack stores compiled slices of the app.
Default modules:
client
– React/redux front-end.db
– core knex database connection helpersrecords
– database record types and repositories, with base record and repository classes inrecord
. Depends ondb
graphql
– Graphql schema and implementation. Depends onrecords
anddb
server
– express.js server that serves the client and graphql api. Depends ongraphql
helpers
– generic helpers that can be used in any other module – no dependencies
This app is set up as a 12-factor app, configurable via environment variables.
The supported environment variables are:
NODE_ENV
–test
,development
, orproduction
DATABASE_URL
– the url of the postgres database.PORT
– port for the server to bind to. Defaults to3001
PUBLIC_HOST
– the public facing domain name to include in e.g. links.REQUIRE_SSL
– if this is notfalse
, all requests are redirected to HTTPS.WEB_CONCURRENCY
– # of workers to use in clustered mode. Clustering disabled if value is 1.NODE_MAX_OLD_SIZE
- limit node process size to a given amount. Defaults to460
MB to work well in 512MB containers, such as heroku.DEV_SERVER_DISABLE_HOST_CHECK
- disables the host check in webpack dev server, to allow testing from a VM or other host.
Run yarn dev
to start up both the server and client at the same time.
yarn dev
runs:
- webpack in watch mode to hot recompile the server
- nodemon to run the server on port
3001
and restart the server on recompilation. - webpack-dev-server to run the client on port
3000
, with proxy through to the server - nodemon processes to regenerate typescript types corresponding to graphql files on change.
The dev server watches for changes and restarts express
each time a dependency file changes.
The dev client is using the webpack-dev-server
to hot reload the client whenever a file changes. Our webpack dev server is configured in webpack/webpack-dev-server.js
.
To build for production, run:
NODE_ENV=production yarn build
This will build the entire app into ./dist/
.
We are using Jest for unit testing, and colocating tests with the modules they’re testing.
Unit tests for a module are located in a __tests__
directory in the same directory as the file being tested. Tests for module.ts
should be named module.test.ts
. Index files should be named after their parent directory. some-module/index.ts
should be tested in some-module/__tests__/module.test.ts
.
To run unit tests, run yarn jest
. This simply runs jest in the current directory, which will use config in the jest
section of package.json
.
To run jest in watch mode, and have it automatically rerun tests when files change:
yarn jest --watch
To see other jest options, you can run:
yarn jest -- --help
yarn test:unit
is an alias for running jest
directly.
We are testing react components with Enzyme . See that for more information.
These tests are included in the Jest (unit) tests (above.)
Acceptance tests are written using Nightmare. See the test:acceptence
tasks for more.
We are using tslint
for linting. It is run automatically before unit tests.
We have experimentally included JSVerify for property-based testing. Property-based testing is based on generating arbitrary inputs for functions and asserting that properties are invariant across those inputs. If an input is found which violates the property, the library will automatically simplify it to the minimal case that reproduces the error.
We are using React Storybook to generate a styleguide for our react components.
You can run the style guide with yarn dev:storybook
We're generating type definitions from our graphql schema, queries, and mutations. This allows us to get static type safety between our graphql code and typescript implementations.
To enable this, we're storing all graphql code in individual .graphql
files. Our build process and dev server look for these and use them to generate the appropriate type definitions.
https://docs.google.com/document/d/1NySKusNwZzChAxY5vnnXnfl9l4JmaGhT3hzyacviuZg/edit
The file modules/graphql/schema-types.ts
is generated by graphql-code-generator
from schema.graphql
and any other .graphql
file in the graphql
module.
schema-types
exports interfaces for all graphql type
s in the schema, including e.g. Query
.
For example, if we have the following schema:
type User {
id: Int!
name: String!
email: String!
}
type Query {
usersById(id: Int!): [User]!
}
schema-types.ts
will contain definitions for:
Query
– containing the return types for each queryUsersByIdQueryArgs
– the expected arguments for theusersById
queryUser
– the straightforward typescript definition forUser
.
Note that we make liberal use of !
in the query definition to disallow null
values as appropriate. !
should not be used when the operation may fail. Graphql
prefers null
returns in that case in most circumstances.
To make use of these types, we import them into our modules/graphql/index.ts
for our resolver definition.
In particular, we would define our usersById
resolver as:
usersById(obj: {}, args: UsersByIdQueryArgs, context: Context): Promise<Query['usersById']> {
...
}
Note that we use UsersByIdQueryArgs
to tell typescript that this should be consistent with the defined schema arguments. We could use an inline type or separate interface, but doing so would defeat TypeScript's ability to tell us when we change the schema that our implementation is no longer compatible.
Similarly, we define the return type to be Promise<Query['usersById']>
. Query['usersById']
is TypeScript syntax that means "whatever tye type of usersById
on in the Query
type is". By using this type subscripting syntax, we get static validation that our resolver is compatible with our schema.
In the client we generate types for our graphql queries and mutations.
Given a .graphql
file containing the query:
query Users($foo: Int!) {
usersById(id: $foo) {
id
name
}
}
Entries will be added to modules/client/graphql-types.ts
:
export interface UsersQueryVariables {
foo: number;
}
export interface UsersQuery {
// Returns all of the users in the system (canned results)
usersById: Array<{
id: number;
name: string;
}>;
}
The types and query can be used with Apollo by require
-ing the .grahql
file directly from typescript and passing it in where a query is expected. The UsersQuery
export async function fetchUsers(id: number): Promise<UsersQuery["usersById"]> {
const vars: UsersQueryVariables = {
foo: id,
};
const result = await graphqlClient.query<UsersQuery>({
query: require("./Answer.graphql"),
variables: vars,
});
return result.data.usersById;
}
The build:graphql
task generates all type files. It:
- Generates
modules/graphql/schema.json
from theschema.graphql
. This is used by subsequent steps. - Generates
schema-types.ts
in the graphql module - Generates
graphql-types.ts
in the client
The dev:graphql
task – which is run automatically by dev
– watches for changes to any .graphql
file and reruns build:graphql
The client has the following capabilities built-in:
- Apollo Client for GraphQL queries/mutations.
- Redux Saga for asynchronous workflows.
- A small lens library for state selectors/updates.
Apollo Client is a smart graphql client with automatic caching and higher-order components for wiring presentation components to graphql queries.
Redux Sagas uses ES7 generator functions to support high level declaration and coordination of asynchronous workflows.
See modules/client/sagas/index.ts
for an example redux saga which uses the apollo client to execute graphql queries.
Accessing/updating functional state in TypeScript requires a different solution from many of the common solutions in JavaScripot – immutable-helper
, for example, is fundamentally untypable.
This starter-kit includes a library of functional lenses which are simple read/write helpers for accessing and updating substate of another object. A lens can be used as a function to get something out of an object, or can have .set
or .update
called on it to create an updated copy of an object.
The lens library is defined in modules/helpers/lenses.ts
. See the tests for examples, as well as use in sagas/index.ts
and reducers/index.ts
.
CSS is implemented using the Trello CSS Guide naming conventions.
We are using the Bourbon stack as our CSS framework.
Instead of one monolithic stylesheet, each component should have its own styles.css
which it requires in it’s main module. This approach eases maintainability, as each react component has its own stylesheet, and webpack will only package up CSS for the components we actually use.
React components should be named with semantic, specific names. However, the CSS classes used for a React component should be as generic as possible.
Prefer using an existing component class name with a new mod-
modifier to specify the behavior of your component versus adding a new css class per react component.
For example, it is better to have one btn
class that has a different mod-foo
modifier for the FooButton
react component, than make a foo-button
class.
See the Trello CSS guide for more info.
The database is postgres and is preconfigured to run in Docker.
To connect via a postgres client:
- Host: 127.0.0.1
- Port: 5432
- Username: root
- No Password
Database names:
-
development
-
test
-
To start:
docker-compose up
and leave running -
To create dev/test databases:
yarn db:create
-
To run psql shell against development DB:
yarn db:run -- development
The project is configured to run in CircleCI (see .circleci/config.yml). It's possible to run locally with circleci local execute
. Warning: the tests may hang after completion if there aren't enough concurrent workers. Fix this by cranking up your cores in Docker or by specifying Jest's --maxWorkers. (We'd prefer not to check in a maxWorkers setting; we want test concurrency to be elastic with the number of hardware threads available.)