Naming Conventions

This document covers naming conventions for the code. Some of these conventions are used by the framework and code generation, and as such are required. Others are our recommendations that we find useful for making the code easier to understand.

Proto files

Proto files are named using the snake_case, as defined by Protobuf. There are several special kinds of files.

identifiers.proto

Commands and events reference model entities using their identifiers. Having typed identifiers makes a model type safe. Although the framework also supports String, Integer, and Long as valid ID types, we strongly recommend defining custom ID types like CustomerId, OrderId, ShipmentId, and others. You can find similar cases in the framework API which has EventId, CommandId, UserId, TenantId, and others.

We find it convenient to define ID types in one file called identifiers.proto. A typical project is likely to have more than one Bounded Context. Thus, you will have several identifiers.proto files. Each of them resides under the directory with proto files defining the data model of the corresponding Bounded Context. For example:


    myproject/
      users/
        src/
          main/
            java/
            proto/
              user.proto
              group.proto
              ...
              identifiers.proto
      tasks/
        src/
          main/
            java/    
            proto/
              task.proto
              project.proto
              ...
              identifiers.proto
        ...

commands.proto

Commands are defined in a file ending with commands.proto. It can be simply commands.proto but usually commands are handled by different entities. Thus, it is convenient to name such a file after the type of the target entity, for example, an aggregate:

  • blog_commands.proto
  • order_commands.proto
  • customer_commands.proto

events.proto

Similarly to commands, events are defined in files which names have the events.proto suffix:

  • blog_events.proto
  • order_events.proto
  • customer_events.proto

rejections.proto

Rejection is a special “negative” kind of events supported by the framework. A rejection is thrown if a command cannot be handled. You may think of them as of exceptions with non-technical flavor.

Similarly to events, rejections are defined in files ending with rejections.proto:

  • blog_rejections.proto
  • order_rejections.proto
  • customer_rejections.proto

For each aggregate you are likely to have all three kinds of files because a command leads to an event, and it is likely there are conditions under which a command cannot be handled.

Entity states

We recommend gathering definition of related entity states in a file named after a business model thing. Suppose we have a Task aggregate, TaskItem and TaskDetails projections, and a Process Manager which is responsible for movement of a task from one project to another, there would be task.proto file, with all Task-related data types definitions. A project-related data types would be defined in a project.proto file.

As it was already mentioned, TaskId and ProjectId are defined in the identifiers.proto file, and task.proto and project.proto import this file.

Data types

Data types are defined as Protobuf messages using TitleCapitalization.

Identifiers

Identifiers are usually defined after the name of the entity with the Id suffix:

  • ProjectId
  • TaskId
  • CommentId

You will find such naming pattern in the framework API. For example, EventId, CommandId, UserId, TenantId, and others.

This convention is not a requirement. We find Id suffix short yet meaningful for building a rich type-safe API. You can also select another convention that fits your domain best. Please note that future version of the framework tools will use the Id suffix of the types for code scaffolding and improving intelligence of code generation.

While the identifier type is usually defined with the Id suffix, we do not recommend following the same strategy for the proto field names. Naming fields as id or adding the _id suffix is usually excessive because the identifier type already has the Id suffix.

Instead, we name the fields by their respective type reference, so user_id becomes user and project_id becomes project.

In events:

message TaskCreated {
    TaskId task = 1;
    ProjectId project = 2;
}

And entity states:

message TaskProjection {
    TaskId task = 1;
    ProjectId project = 2
    string name = 3;
}

The only exception from the suggestion is when the ID is a part of the root aggregate state, or a command that creates the aggregate directly.

In aggregates:

message Task {
    TaskId id = 1;
    ProjectId project = 2;
}

And entity creation commands:

message CreateTask {
    TaskId id = 1;
    ProjectId project = 2;
}

repeated and map fields

We recommend naming repeated and map fields using singular nouns as such a naming appears to be closer to the language we speak. It also provides easier to use generated code.

This convention contradicts with the official Protobuf Style Guide which suggests naming map and repeated fields after plural nouns. Knowing this, we still recommend singular because of the following. The code generated for a repeated and map field named after a singular noun is closer to real English. For the code related to the Domain-Driven Design this is far more important than the consistency with the style guide.

So when defining repeated and map fields use:

Singulars

message Task {
    TaskId id = 1;
    repeated SubTaskId subtask = 2;
    map<string, string> user_label = 3;
}

Over pluralized names

message CreateTask {
    TaskId id = 1;
    repeated SubTaskId subtasks = 2;
    map<string, string> user_labels = 3;
}

Commands

A command is defined as an imperative:

  • CreateProject
  • AssignTask
  • RemoveComment

Events

Events are named as facts formulated as past participles, for example:

  • ProjectCreated
  • TaskAssigned
  • CommentRemoved

Rejections

A rejection is named after a reason of why a command cannot be handled. In fact, rejection notifies on a state of the domain model, for example:

  • TaskAlreadyExists
  • InsufficientFunds
  • ProjectAlreadyCompleted

Entity states

Protobuf messages for entity states are defined using nouns, for example:

  • Project
  • Task
  • Comment

Avoid using suffixes like Aggregate, Projection, ProcessManager when defining a proto type for the following reasons:

  1. You may want to use such a word when creating an entity Java class which uses a generated data type for holding the state of the entity.
  2. Such data structure does not represent a whole Aggregate or ProcessManager thing anyway. It is just data.

Packages

Packages allow to form namespaces for types and avoid clashes. It is customary to have a “root” package for an organization or a service name. Most likely each Bounded Context would have a dedicated package.

Examples in this guide assume that a fictitious company called Acme Corporation creates a SaaS solution. The company has a web presence site with the domain name acme.io. The solution is a task management application called “Todo List” which will be hosted at todolist.acme.io.

Proto packages

Packages in Protobuf do not follow the reverse Internet domain name convention, which is customary in Java. It would make sense to have a root package for all types defined in an organization under the root package with a lowercase company name.

For the fictitious SaaS project of the Acme Corporation it would be:

package acme.todolist;

Java packages

Java does not have the notion of package nesting. Packages in Java are separated namespaces, which seem hierarchical for convenience. When it comes to placing source code files in a project, there is usually nesting formed by the directories in a file system.

Spine framework uses this notion of “nesting” for marking multiple packages of a server-side code belonging to a Bounded Context easier. But this is a convenience feature, not a requirement.

Please see our recommendations for organizing generated and handcrafted code in sections below.

Have package per data type

It is easier to see everything related to a type, if all the generated code comes under a “home” package of a data type. For example:

  • io.acme.todolist.task
  • io.acme.todolist.project
  • io.acme.todolist.comment

This package would be a part of API shared between client- and server-side code of your application.

Commands

We recommend putting command classes under a package which ends with command:

  • io.acme.todolist.task.command
  • io.acme.todolist.project.command
  • io.acme.todolist.comment.command

The package name is singular because it reads better in a fully-qualified class name of a command message.

Events

Similarly to commands, we recommend putting events generated by an entity under the entity sub-package:

  • io.acme.todolist.task.event
  • io.acme.todolist.project.event
  • io.acme.todolist.comment.event

Rejections

Similarly to events, rejections are placed under the package called rejection:

  • io.acme.todolist.task.rejection
  • io.acme.todolist.project.rejection
  • io.acme.todolist.comment.rejection

Unlike commands and events, rejection messages are generated under a file named Rejections. The class is placed into a rejection package of the corresponding type.

The package also contains generated Throwable top-level classes that match rejection messages. These classes are used in the throws clause of command handling methods.

The arrangement with message classes nested under Rejections class and top-level Throwables is required to avoid name clashes while keeping these generated classes under the same package.

For details on rejections usage, refer to Defining Rejections Guide.

Server-side code

To avoid unwanted dependencies we find it useful to put server-side code under a sub-package called server with sub-packages for corresponding entity types:

  • io.acme.todolist.server.task
  • io.acme.todolist.server.project
  • io.acme.todolist.server.comment

Handcrafted Java Classes

Entities

When naming entities we find it natural to start with a name of a state class and then add a suffix which tells the type of the entity:

  • ProjectAggregate
  • OrderProcessManager
  • TaskItemProjection

The suffix helps for observing together with other entities in a package.

For process managers it may be enough to have the Process suffix dropping Manager which frequently worked for us too. Other options for suffixes are Pm or Procman.

It would be a good idea to decide on such suffix as a team standard before you start coding.

Repositories

We recommend not using a type infix for naming repository classes. Alphabetical sorting would make a repository class be next to an entity class, and you would not deal much with repository classes anyway. Thus, it is just SomethingRepository rather than SomethingAggregateRepository:

  • ProjectRepository
  • OrderRepository
  • TaskItemRepository

Bounded Contexts

Names

Bounded Contexts names follow TitleCapitalization favoring plurals:

  • Users
  • Tasks
  • DeliveredOrders

Although, singular names are perfectly fine too, for example:

  • Billing
  • Shipping
  • DynamiteProduction

Packages

If a name of a Bounded Context is used in a package, its name is transformed according to the rules of a programming language.

Factory classes

A Java class that creates and configures an instance of a BoundedContext is named after the name of the context with the Context prefix:

  • UsersContext
  • DeliveredOrdersContext
  • OnboardingContext