📎 Webclip
Dependency Injection in Crystal
The article presents dependency injection as a way to make Crystal applications easier to test, more flexible, and less tied to concrete classes. It walks through three uses: managing multiple partner instances, swapping writer implementations through abstractions, and sharing current-user data across services in the same request.
Reading notes#
- DI is described as a tool for making applications easier to test, more expandable, and more flexible.
- The article uses Crystal and Athena, and notes that Athena’s DI component is also available as a standalone shard.
- In the manager example, a
PartnerManagerresolves a partner ID into aPartnerinstance, which lets new partners be added by registering another service. - The manager approach keeps partner-specific logic out of the controller and avoids hard-coding an array of partner objects.
- In the plug-and-play example, a
Workerdepends on aWriterInterfaceinstead of a concrete writer class. - When there are multiple writers, the article shows two resolution options: a service alias for the default implementation or the initializer argument name to pick a specific service.
- DI is used to support testing by injecting a mock writer into
Workerso the class can be tested in isolation. - The article says this approach lets services depend on abstractions and makes it easier to reuse services that have a single responsibility.
- In the sharing-data example, a
UserStorageservice holds the current user, and aSecurityListenersets that user during request handling. - The listener also adds log context with the user’s information, so later logs in the same fiber include that data.
- The article notes that the DI container is fiber-specific, which makes per-request sharing possible without manual cleanup.
- It contrasts this with storing
current_useronHTTP::Server::Context, using a class variable, or storing the user in a database, and argues those options are less flexible in some scenarios. - The conclusion says DI reduces coupling, improves testing, and encourages reuse, but should be applied where the problem fits.
