OpenProject development style guide - frontend

Code format

OpenProject follows the AirBnB’s style guide regarding to the code format.

Development patterns

Angularized

OpenProject follows the Angular’s style guide patterns.

Declarative

Declarative Programming is a paradigm where the code describes what to do by encapsulating the how to do it (implementation details) under abstractions. The ultimate result of declarative programming is the creation of a new Domain Specific Language (DSL).

What

Encapsulate logic in methods with meaningful names.
- Domain/Business logic:
  Encapsulate the implementation details in the subdomain service (e.g., login should be placed in the AuthService and encapsulate all the login functionality).
- Presentation or interaction logic:
  Encapsulate the implementation details in a component’s method or in a Presenter (service scoped/provided in the component).

Example

// Imperative programming
const bestProducts = [];

for(let i = 0; i < products.length; i++) {
  let product = products[i];
  if (product.rating >= 5 && product.price < 100) {
     bestProducts.push(product);
  }
}

// More declarative
const bestProducts = products.filter(function(product) {
 return product.rating >= 5 && product.price < 100;
});

// Most declarative, implementation details are hidden in a function
const bestProducts = getBestProducts();

An example in OpenProject would be the APIV3Service that encapsulates all the logic to deal with the OpenProject API.

Why

Code is easier to understand and reason about.
Favours reusability.
Simplifies refactoring.
The state management becomes predictable, easy to trace, and under control.
Aligns with unidirectional data flow (Stores) and the Components Architecture.
Increases coding pleasure and productivity.

Immutable

Not capable of or susceptible to change. An immutable value can’t be changed, when edited a new copy is returned.

What

Do not mutate objects, spread the word.

Do not edit object’s, use the immutable alternatives:

const copy = {...originalObject}; 
const add = {...originalObject, propertyToChange: 'new value'};
const remove = {propertyToDelete, ...newObjectWithoutThePropertyToDelete};

Avoid Array mutation methods (push, pop, shift, unshift, sort, reverse, splice, delete), use their immutable alternatives:

clone = x => [...x];
push = y => x => [...x, y];
pop = x => x.slice(0, -1);
unshift = y => x => [y, ...x];
shift = x => x.slice(1);
sort = f => x => [...x].sort(f);
delete = i => x => [...x.slice(0, i), ...x.slice(i + 1)];
splice = (s, c, ...y) => x => [...x.slice(0, s), ...y, ...x.slice(s + c)];

Why

In Javascript, Objects can be edited by reference, opening the door to unpredictable mutations that can have unintended effects on other parts of the app.
Makes code easier to understand because data changes become explicit and obvious.
Avoids a set of hard to detect bugs.
Enables unidirectional data flow.
Enables performance improvements through ChangeDetectionStrategy.OnPush

Unidirectional data flow

The app has a single way to read and to write the state, and both are separated (Command Query Segregation).

State definition

We can differentiate two types of states in our application:

Local: belongs to a single component. Includes mainly UI state.
Global: belongs the full app. This includes a shared representation of the backend queries and models, but also things like the current Route, logged in user and capabilities of this user.

There are also two types of other state that our frontend application has to be concerned with:

Global Remote: the state of the server.
Local Remote: the state of other clients.

For the server, most of the time this means the database contents. Keeping our local state up-to-date with this is a hard problem to solve, and will force us to make unwanted API calls or unprovable assertions. Long-term, we would like to have live (e.g. websocket) connections for pieces of state, so we can always be up-to-date with the database without bombarding the server in requests.

Syncing with remote clients’ state is currently nonexistent. Clients submit their updates only to the server. When submitting, clients submit a version token of the resource, which gets updated by the server. If a client tries to work on an older version of the resource, the request will fail. Usually is no sophisticated method in place to handle these; an error toaster will be thrown. Direct state transfer or sharing between clients is currently not done. Since this is a very hard problem to solve, there are no plans to change this significantly.

State relating to the component’s view (or children) should be declared and managed in that component. Children that rely on this state should receive it via inputs and request changes via outputs. Sometimes, this tree might prove too complex to easily hand local state and events up and down via this mechanism. In that case you may create an Akita store that is injected into the first shared parent component, and manage the state there. You must not save global state in a local component or service. You should not save state in non-akita services. The goal is to have a unified, observable-based interface to all application state.

Most of our backend-related data is in the entity format. To capture this, there must be a global entity store. An example implementation is the in-app-notification store. This store olds a reference of all entities of a particular type that are in-use somewhere in the application as well as a list of IDs for entity collections of that type. Stores and components consuming a particular entity type must go through the global entity store to perform CRUD operations, so that updates can be properly reflected across the application.

Events and side effects

Mutable operations on the entities can have side effects on different collections and entities currently in use by other parts of the application. Oftentimes, the frontend cannot know beforehand which operations will have what kind of impact. This means that the respective collections and entities have to be refreshed from the backend. Some examples:

Marking a notification as read in the work package details tab should update the counter of the notification bell in the header
Changing the type of a work package in the split view changes the collection that is shown in the table, since the work package will be filtered out.

For this use-case, we have implemented a global actions service. You can dispatch actions here, and other parts of the application can listen to these actions. Think of it like a global event bus. These actions are typed.

To reduce server requests, side effects should be be calculated in the frontend. If this is impossible, the updating store must send out a global event to notify other parts that the specific event occurred.

Note: The proper solution to this problem would be a backend that can push updates for collections and entities that we are requiring. However, implementing and relying on websockets comes with its own challenges.

Flow

data-flow

Component/Service requests an update to the Store (directly or through an Action)
Store changes the state and emits an updated copy of the state
Component receives the update and renders the new state

Unidirectional data flow in the view / component’s tree

Angular also follows the unidirectional data flow pattern in the view to improve the performance and simplify the state distribution:

When the state could have been updated (an async operation happens), Angular performs Change Detection in one way, from top to bottom, from the root component all the tree down checking every child component for changes.
Every component that receives state updates (@Input, Services, UI…), updates itself, renders its view and then triggers Change Detection for its child components.
If the children would trigger state changes up to the parent (bidirectional data flow (two way data binding)), the parent should update itself and then trigger Change Detection for the children, which could trigger changes up to the parent… causing a loop.
This is why unidirectional data flow is enforced in development mode (ExpressionChangedAfterItHasBeenCheckedError) and encouraged in the Components Architecture.

What

Do use the chosen state management library by default.

Why

Stores concentrate the state management, allowing the different parts of the application to remain highly decoupled.
Container components concentrate the state distribution through @Input and @Outputs, so the state flow is clear just by taking a look at the template.
The state management becomes predictable, easy to trace, and under control.
Reduces side effects and facilitates debugging.
Encourages data normalization and avoids duplications.
Code is easier to understand and reason about.
Increases coding pleasure and productivity.

Components architecture

Mental mindset to build clearer apps based on the differentiation between Container Components (CC), Presentational Components (PC) and Services.

Services

Are state and logic containers.

What

Encapsulate logic in services:
- State (Store) + Domain logic (state CRUD)
- Complex UI logic (Presenters)
Provide a meaningful API to access the state and logic.
Follow the unidirectional data flow pattern

Why

Decouples state management from components.
Allows to keep the components lean.
Scoped Shareability: allow to share the state and logic scoped to components, modules or globally.
Testability
Encourages Declarative Programming
Testability: easier to test because there are no side effects (pure).

Container components (CC)

Represent a feature that interacts with the state. This could be a page (routed component) but also standalone components (e.g. sign-in button (tied to the AuthService)).

Responsibilities

Inject the State Services (Stores).
Contain Presentational Components (PC), usually.
Bridge State Services (Stores) and PCs:
- Pass down state updates (from Stores) through PCs’ @Outputs
- React to PCs’ @Outputs
Notes:
- Could contain other CCs.
- Could contain presentational logic (show/hide PCs, calculations..), in those cases are named something like Mixed Components.

Example

Pages, components that are routed, are usually container components since the need to fetch state to display it.

What

Concentrate the interaction with the state (Stores) in Container Components.

Why

Separation of concerns; separates state interaction from UI (presentation and interaction).
Concentrates state interaction; the state management becomes more predictable, easy to trace, and under control.
Simplifies the state flow: the state distribution through @Input and @Outputs makes the state flow clear just by taking a look at the template.
Reduces side effects and facilitates debugging.
Code is easier to understand and reason about.
Increases coding pleasure and productivity.

Presentational components

Are the building blocks of the UI.

Responsibilities

Present/display state that is passed via @Inputs.
Handle user interaction:
- Implement the interaction logic
- Emit @Outputs
Notes:
- Could contain:
  - Other Presentational Components and Container Components.
  - Presenter Services that encapsulate complex UI logic (computed styles, UI calculations…)).

Example

Components from UI libraries are usually Presentational Components (e.g., material button).

What

Create presentational components to encapsulate UI logic (presentation or interaction).

Why

Reusability:
Presentational components are more reusable because:
- Are not tied to a concrete business logic.
- Usually perform a generic interaction/presentation (button, tab, list, layout)
Increase
- Productivity
- Standardization
Align with Design Systems
Improved testability because there are no side effects (pure).
Are easier to replace.