- Local Development
- Running The Tests
- Debugging
- Structure of the app
- Philosophy
- Internationalization
- Deploying
These instructions should get you set up ready to work on New Expensify 🙌
- Install
node
&npm
:brew install node
- Install
watchman
:brew install watchman
- Install dependencies:
npm install
You can use any IDE or code editing tool for developing on any platform. Use your favorite!
- To run the development web app:
npm run web
- Changes applied to Javascript will be applied automatically via WebPack as configured in
webpack.dev.js
- To install the iOS dependencies, run:
npm install && cd ios/ && pod install && cd ..
- To run a on a Development Simulator:
npm run ios
- Changes applied to Javascript will be applied automatically, any changes to native code will require a recompile
- To install the Android dependencies, run:
npm install
- Make sure you have Java installed
java -version
. If not, install it by runningnpm install -g openjdk8
. - To run a on a Development Emulator:
npm run android
- Changes applied to Javascript will be applied automatically, any changes to native code will require a recompile
- To run the Development app, run:
npm run desktop
, this will start a new Electron process running on your MacOS desktop in thedist/Mac
folder.
- If you are having issues with Getting Started, please reference React Native's Documentation
- If you are running into issues communicating with the API please verify your
.env
file is set up correctly for the platform you are trying to run.
Note: Expensify engineers that will be testing with the API in your local dev environment please refer to these additional instructions.
Unit tests are valuable when you want to test one component. They should be short, fast, and ideally only test one thing. Often times in order to write a unit test, you may need to mock data, a component, or library. We use the library Jest to help run our Unit tests.
- To run the Jest unit tests:
npm run test
End to end tests are valuable when we do not want to mock data and run against the actual compiled app on iOS or Android. In order to run the end to end tests, we have to compile the iOS or Android app, then launch a simulator, then run tests. We use Detox a "Gray box end-to-end testing and automation library" to help with our end to end testing.
You are first required to build the tests, then you can run them:
- To build the Detox end to end tests:
npm run detox-build
- To run the Detox end to end tests:
npm run detox-test
- If running on the iOS simulator pressing
⌘D
will open the debugging menu. - This will allow you to attach a debugger in your IDE, React Developer Tools, or your browser.
- For more information on how to attach a debugger, see React Native Debugging Documentation
Our React Native Android app now uses the Hermes
JS engine which requires your browser for remote debugging. These instructions are specific to Chrome since that's what the Hermes documentation provided.
- Navigate to
chrome://inspect
- Use the
Configure...
button to add the Metro server address (typicallylocalhost:8081
, check yourMetro
output) - You should now see a "Hermes React Native" target with an "inspect" link which can be used to bring up a debugger. If you don't see the "inspect" link, make sure the Metro server is running.
- You can now use the Chrome debug tools. See React Native Debugging Hermes
These are the main pieces of the application.
This is a persistent storage solution wrapped in a Pub/Sub library. In general that means:
- Onyx stores and retrieves data from persistent storage
- Data is stored as key/value pairs, where the value can be anything from a single piece of data to a complex object
- Collections of data are usually not stored as a single key (eg. an array with multiple objects), but as individual keys+ID (eg.
report_1234
,report_4567
, etc.). Store collections as individual keys when a component will bind directly to one of those keys. For example: reports are stored as individual keys becauseOptionRow.js
binds to the individual report keys for each link. However, report actions are stored as an array of objects because nothing binds directly to a single report action. - Onyx allows other code to subscribe to changes in data, and then publishes change events whenever data is changed
- Anything needing to read Onyx data needs to:
- Know what key the data is stored in (for web, you can find this by looking in the JS console > Application > local storage)
- Subscribe to changes of the data for a particular key or set of keys. React components use
withOnyx()
and non-React libs useOnyx.connect()
. - Get initialized with the current value of that key from persistent storage (Onyx does this by calling
setState()
or triggering thecallback
with the values currently on disk as part of the connection process)
- Subscribing to Onyx keys is done using a constant defined in
ONYXKEYS
. Each Onyx key represents either a collection of items or a specific entry in storage. For example, since all reports are stored as individual keys likereport_1234
, if code needs to know about all the reports (eg. display a list of them in the nav menu), then it would subscribe to the keyONYXKEYS.COLLECTION.REPORT
.
Actions are responsible for managing what is on disk. This is usually:
- Subscribing to Pusher events to receive data from the server that will get put immediately into Onyx
- Making XHRs to request necessary data from the server and then immediately putting that data into Onyx
- Handling any business logic with input coming from the UI layer
This layer is solely responsible for:
- Reflecting exactly the data that is in persistent storage by using
withOnyx()
to bind to Onyx data. - Taking user input and passing it to an action
Almost all the code is located in the src
folder, inside it there's some organization, we chose to name directories that are
created to house a collection of items in plural form and using camelCase (eg: pages, libs, etc), the main ones we have for now are:
- components: React native components that are re-used in several places.
- libs: Library classes/functions, these are not React native components (ie: they are not UI)
- pages: These are components that define pages in the app. The component that defines the page itself should be named
<pageName>Page
if there are components used only inside one page, they should live in its own directory named after the<pageName>
. - styles: These files define styles used among components/pages
Files should be named after the component/function/constants they export, respecting the casing used for it. ie:
- If you export a constant named
CONST
it's file/directory should be named theCONST
. - If you export a component named
Text
the file/directory should be namedText
- If you export a function named
guid
the file/directory should be namedguid
. - For files that are utilities that export several functions/classes use the UpperCamelCase version ie:
DateUtils
. - HOCs should be named in camelCase like withOnyx.
- All React components should be PascalCase (a.k.a. UpperCamelCase 🐫).
In most cases, the code written for this repo should be platform-independent. In such cases, each module should have a single file, index.js
, which defines the module's exports. There are, however, some cases in which a feature is intrinsically tied to the underlying platform. In such cases, the following file extensions can be used to export platform-specific code from a module:
- Mobile =>
index.native.js
- iOS/Android =>
index.ios.js
/index.android.js
- Web =>
index.website.js
- Desktop =>
index.desktop.js
Note that index.js
should be the default and only platform-specific implementations should be done in their respective files. i.e: If you have mobile-specific implementation in index.native.js
, then the desktop/web implementation can be contained in a shared index.js
. Furthermore, index.native.js
should not be included in the same module as index.ios.js
or index.android.js
.
When adding new API commands (and preferably when starting using a new one that was not yet used in this codebase) always
prefer to return the created/updated data in the command itself, instead of saving and reloading. ie: if we call CreateTransaction
,
we should prefer making CreateTransaction
return the data it just created instead of calling CreateTransaction
then Get
rvl=transactionList
Different platforms come with varying storage capacities and Onyx has a way to gracefully fail when those storage limits are encountered. When Onyx fails to set or modify a key the following steps are taken:
- Onyx looks at a list of recently accessed keys (access is defined as subscribed to or modified) and locates the key that was least recently accessed
- It then deletes this key and retries the original operation
By default, Onyx will not evict anything from storage and will presume all keys are "unsafe" to remove unless explicitly told otherwise.
To flag a key as safe for removal:
- Add the key to the
safeEvictionKeys
option inOnyx.init(options)
- Implement
canEvict
in the Onyx config for each component subscribing to a key - The key will only be deleted when all subscribers return
true
forcanEvict
e.g.
Onyx.init({
safeEvictionKeys: [ONYXKEYS.COLLECTION.REPORT_ACTIONS],
});
export default withOnyx({
reportActions: {
key: ({reportID}) => `${ONYXKEYS.COLLECTION.REPORT_ACTIONS}${reportID}`,
canEvict: props => !props.isActiveReport,
},
})(ReportActionsView);
- The major difference between React-Native and React are the components that are used in the
render()
method. Everything else is exactly the same. If you learn React, you've already learned 98% of React-Native. - The application uses React-Router for navigating between parts of the app.
- Higher Order Components are used to connect React components to persistent storage via Onyx.
This application is built with the following principles.
-
Data Flow - Ideally, this is how data flows through the app:
- Server pushes data to the disk of any client (Server -> Pusher event -> Action listening to pusher event -> Onyx). Currently the code only does this with report comments. Until we make more server changes, this steps is actually done by the client requesting data from the server via XHR and then storing the response in Onyx.
- Disk pushes data to the UI (Onyx -> withOnyx()/connect() -> React component).
- UI pushes data to people's brains (React component -> device screen).
- Brain pushes data into UI inputs (Device input -> React component).
- UI inputs push data to the server (React component -> Action -> XHR to server).
- Go to 1
-
Offline first
- All data that is brought into the app and is necessary to display the app when offline should be stored on disk in persistent storage (eg. localStorage on browser platforms). AsyncStorage is a cross-platform abstraction layer that is used to access persistent storage.
- All data that is displayed, comes from persistent storage.
-
UI Binds to data on disk
- Onyx is a Pub/Sub library to connect the application to the data stored on disk.
- UI components subscribe to Onyx (using
withOnyx()
) and any change to the Onyx data is published to the component by callingsetState()
with the changed data. - Libraries subscribe to Onyx (with
Onyx.connect()
) and any change to the Onyx data is published to the callback with the changed data. - The UI should never call any Onyx methods except for
Onyx.connect()
. That is the job of Actions (see next section). - The UI always triggers an Action when something needs to happen (eg. a person inputs data, the UI triggers an Action with this data).
- The UI should be as flexible as possible when it comes to:
- Incomplete or missing data. Always assume data is incomplete or not there. For example, when a comment is pushed to the client from a pusher event, it's possible that Onyx does not have data for that report yet. That's OK. A partial report object is added to Onyx for the report key
report_1234 = {reportID: 1234, isUnread: true}
. Then there is code that monitors Onyx for reports with incomplete data, and callsfetchChatReportsByIDs(1234)
to get the full data for that report. The UI should be able to gracefully handle the report object not being complete. In this example, the sidebar wouldn't display any report that doesn't have a report name. - The order that actions are done in. All actions should be done in parallel instead of sequence.
- Parallel actions are asynchronous methods that don't return promises. Any number of these actions can be called at one time and it doesn't matter what order they happen in or when they complete.
- In-Sequence actions are asynchronous methods that return promises. This is necessary when one asynchronous method depends on the results from a previous asynchronous method. Example: Making an XHR to
command=CreateChatReport
which returns a reportID which is used to callcommand=Get&rvl=reportStuff
.
- Incomplete or missing data. Always assume data is incomplete or not there. For example, when a comment is pushed to the client from a pusher event, it's possible that Onyx does not have data for that report yet. That's OK. A partial report object is added to Onyx for the report key
-
Actions manage Onyx Data
- When data needs to be written to or read from the server, this is done through Actions only.
- Public action methods should never return anything (not data or a promise). This is done to ensure that action methods can be called in parallel with no dependency on other methods (see discussion above).
- Actions should favor using
Onyx.merge()
overOnyx.set()
so that other values in an object aren't completely overwritten. - In general, the operations that happen inside an action should be done in parallel and not in sequence (eg. don't use the promise of one Onyx method to trigger a second Onyx method). Onyx is built so that every operation is done in parallel and it doesn't matter what order they finish in. XHRs on the other hand need to be handled in sequence with promise chains in order to access and act upon the response.
- If an Action needs to access data stored on disk, use a local variable and
Onyx.connect()
- Data should be optimistically stored on disk whenever possible without waiting for a server response. Example of creating a new optimistic comment:
- user adds a comment
- comment is shown in the UI (by mocking the expected response from the server)
- comment is created in the server
- server responds
- UI updates with data from the server
-
Cross Platform 99.9999%
- A feature isn't done until it works on all platforms. Accordingly, don't even bother writing a platform-specific code block because you're just going to need to undo it.
- If the reason you can't write cross platform code is because there is a bug in ReactNative that is preventing it from working, the correct action is to fix RN and submit a PR upstream -- not to hack around RN bugs with platform-specific code paths.
- If there is a feature that simply doesn't exist on all platforms and thus doesn't exist in RN, rather than doing if (platform=iOS) { }, instead write a "shim" library that is implemented with NOOPs on the other platforms. For example, rather than injecting platform-specific multi-tab code (which can only work on browsers, because it's the only platform with multiple tabs), write a TabManager class that just is NOOP for non-browser platforms. This encapsulates the platform-specific code into a platform library, rather than sprinkling through the business logic.
- Put all platform specific code in dedicated files and folders, like /platform, and reject any PR that attempts to put platform-specific code anywhere else. This maintains a strict separation between business logic and platform code.
This application is built with Internationalization (I18n) / Localization (L10n) support, so it's important to always localize the following types of data when presented to the user (even accessibility texts that are not rendered):
- Texts: See translate method
- Date/time: see DateUtils
- Numbers and amounts: see numberFormat
- Phones: see LocalPhoneNumber
In most cases, you will be needing to localize data used in a component, if that's the case, there's a HOC withLocalize. It will abstract most of the logic you need (mostly subscribe to the NVP_PREFERRED_LOCALE Onyx key) and is the preferred way of localizing things inside components.
Some pointers:
- All translations are stored in language files in src/languages.
- We try to group translations by their pages/components
- A common rule of thumb is to move a common word/phrase to be shared when it's in 3 places
- Always prefer longer and more complex strings in the translation files. For example
if you need to generate the text
User has sent $20.00 to you on Oct 25th at 10:05am
, add just one key to the translation file and use the arrow function version, like so:nameOfTheKey: ({amount, dateTime}) => "User has sent " + amount + " to you on " + dateTime,
. This is because the order of the phrases might vary from one language to another.
We utilize a CI/CD deployment system built using GitHub Actions to ensure that new code is automatically deployed to our users as fast as possible. As part of this process, all code is first deployed to our staging environments, where it undergoes quality assurance (QA) testing before it is deployed to production. Typically, pull requests are deployed to staging immediately after they are merged.
Every time a PR is deployed to staging, it is added to a special tracking issue with the label StagingDeployCash
(there will only ever be one open at a time). This tracking issue contains information about the new application version, a list of recently deployed pull requests, and any issues found on staging that are not present on production. Every weekday at 9am PST, our QA team adds the 🔐LockCashDeploys🔐
label to that tracking issue, and that signifies that they are starting their daily QA cycle. They will perform both regular regression testing and the QA steps listed for every pull request on the StagingDeployCash
checklist.
Once the StagingDeployCash
is locked, we won't run any staging deploys until it is either unlocked, or we run a production deploy. If severe issues are found on staging that are not present on production, a new issue (or the PR that caused the issue) will be labeled with DeployBlockerCash
, and added to the StagingDeployCash
deploy checklist. If we want to resolve a deploy blocker by reverting a pull request or deploying a hotfix directly to the staging environment, we can merge a pull request with the CP Staging
label.
Once we have confirmed to the best of our ability that there are no deploy-blocking issues and that all our new features are working as expected on staging, we'll close the StagingDeployCash
. That will automatically trigger a production deployment, open a new StagingDeployCash
checklist, and deploy to staging any pull requests that were merged while the previous checklist was locked.
These are some of the most central GitHub Workflows. There is more detailed information in the README here.
The preDeploy workflow executes whenever a pull request is merged to main
, and at a high level does the following:
- If the
StagingDeployCash
is locked, comment on the merged PR that it will be deployed later. - Otherwise:
- Create a new version by triggering the
createNewVersion
workflow - Use the
updateProtectedBranch
workflow to update thestaging
branch.
- Create a new version by triggering the
- Also, if the pull request has the
CP Staging
label, it will execute thecherryPick
workflow to deploy the pull request directly to staging, even if theStagingDeployCash
is locked
The deploy
workflow is really quite simple. It runs when code is pushed to the staging
or production
branches, and:
- If
staging
was updated, it creates a tag matching the new version, and pushes tags. - If
production
was updated, it creates a GitHub Release for the new version
The platformDeploy
workflow is what actually runs the deployment on all four platforms (iOS, Android, Web, macOS Desktop). It runs a staging deploy whenever a new tag is pushed to GitHub, and runs a production deploy whenever a new release is created.
The lockDeploys
workflow executes when the StagingDeployCash
is locked, and it prepares the staging branch for a production release by creating a new PATCH
version (i.e: 1.0.57-5
-> 1.0.58.0
).
The finishReleaseCycle
workflow executes when the StagingDeployCash
is closed. It updates the production
branch from staging
(triggering a production deploy), deploys main
to staging, and creates a new StagingDeployCash
deploy checklist.
Sometimes it might be beneficial to generate a local production version instead of testing on production. Follow the steps below for each client:
In order to generate a production web build, run npm run build
, this will generate a production javascript build in the dist/
folder.
In order to compile a production desktop build, run npm run desktop-build
, this will generate a production app in the dist/Mac
folder named Chat.app
.
In order to compile a production iOS build, run npm run ios-build
, this will generate a Chat.ipa
in the root directory of this project.
To build an APK to share run (e.g. via Slack), run npm run android-build
, this will generate a new APK in the android/app
folder.