coding-principles.md

Coding Principles and Style Guide

Fail-fast and Explicit Configuration

1. Configuration should be loaded and validated at extension initialization so that issues are reported immediately. Do not lazy-load configuration unless it is required to do so.
2. Settings can be pulled from the extension context and placed into configuration objects, which are passed to services via their constructor.
3. Service configuration requirements should always be explicit; as a general rule, do not pass a single configuration object with many values to multiple services.
4. Annotate configuration keys with @Setting so that they may be tracked.

Errors

1. Do not throw checked exceptions; always use unchecked exceptions. If an unchecked exception type needs to be defined, inherit from org.eclipse.edc.spi.EdcException.
2. Do not throw exceptions to signal a validation error; report the error (preferably collated) and return an error response.
3. Throw an unchecked exception if something unexpected happens (e.g. a backing store connection is down after a number of retries). Note that validation errors are expected.
4. Only throw an exception when there is no remediation possible, i.e. the exception is fatal. Do not throw an exception if an operation can be retried.
5. Generally, services should return a org.eclipse.edc.spi.result.Result (or a suitable subclass).

Simplicity

1. Avoid layers of indirection when they are not needed (e.g. "pass-through methods").
2. Avoid needlessly wrapping objects, especially primitive datatypes.

General Coding Style

1. Use var instead of explicit types (helps with clarity)
2. Avoid final in method args and local variables
3. Use final in field declarations
4. Avoid static fields except in constants or when absolutely necessary (you should be able to provide a reason).
5. Use interfaces to define shared constants
6. Use "minimally required types" (or "smallest possible API"), e.g. use ObjectMapper instead of TypeManager, or use a String instead of a more complex object containing the String, etc.
7. Use either public members, which are documented and tested, or private members.
8. Avoid package-private members, especially if only needed for testing
9. Avoid protected members unless they're intended to be overridden.
10. Use package-private classes if they're not needed outside the package, e.g. implementation classes
11. Avoid using enums for anything other than named integer enumerations.
12. Avoid using static classes as much as possible. Exceptions to this are helper functions and test utils, etc. as well as static inner classes.
13. Use only camel case and no prefixes for naming.
14. Avoid unnecessary this. except when it is necessary e.g. when there is a name overlap
15. Use static imports, as long as code readability and comprehension is not impacted. For example,
    • use assertThat(...) instead of Assertions.assertThat(...)
    • use format("...",arg1) instead of String.format(...), but
    • avoid of(item1, item2).map(it -> it.someOperation)... instead of Stream.of(item1, item2). Also, avoid static imports if two static methods with the same name would be imported from different classes
16. Avoid Optional as method return type or method argument, except when designing a fluent API. Use null in signatures.
17. Avoid cryptic variable names, especially in long methods. Instead, try to write them out, at least to a reasonable extent.

Testing

1. All handlers and services should have dedicated unit tests with mocks used for dependencies.
2. Prefer unit tests over all other test types: unit > integration/component > e2e
3. When appropriate, prefer composing services via the constructor so that dependencies can be mocked as opposed to instantiating dependencies directly.
4. Use classes with static test functions to provide common helper methods, e.g. to instantiate an object.
5. Use [METHOD]_when[CONDITION]_should[EXPECTATION] as naming template for test methods, e.g. verifyInput_whenNull_shouldThrowNpe() as opposed to testInputNull()

Data Objects

1. Use the Builder pattern when:
   • there are any number of optional constructor args
   • there are more than 3 constructor args
   • inheriting from an object that fulfills any of the above. In this case use derived builders as well.
2. Although serializability is not the reason we use the builder pattern, it is a strong indication that a builder should be used.
3. Builders should be named just Builder and be static nested classes.
4. Create a public static Builder newInstance(){...} method to instantiate the builder
5. Builders have non-public constructors
6. Use single-field builders: a Builder instantiates the object it builds in its constructor, and sets the properties in its builder methods. The build() method then only performs verification (optional) and returns the instance.
7. Use private constructors for the objects that the builder builds.
8. If there is a builder for an object, use it to deserialize an object, i.e. put Jackson annotations such as JsonCreator and @JsonBuilder on builders.
9. Note that the motivation behind use of builders is not for immutability (although that may be good in certain circumstances). Rather, it is to make code less error-prone and simpler given the lack of named arguments and optional parameters in Java.

Secrets

1. Only store secrets in the Vault and do not hold them in objects that may be persisted to other stores.
2. Do not log secrets or sensitive information.

Extensions and Libraries

1. Extension modules contribute a feature to the runtime such as a service.
2. SPI modules define extensibility points in the runtime. There is a core SPI module that defines extensibility for essential runtime features. There are other SPI modules that define extensibility points for optional features such as IDS.
3. Libraries are utility modules that provide classes which may be used by other modules. They do not directly contribute features to the runtime.
4. An SPI module may only reference other SPI modules and library modules.
5. An Extension module may only reference other SPI modules and library modules.
6. A library module may only reference other library modules.

Build

1. There should only be a root gradle.properties that contains build variables. Do not create separate gradle.properties files in a module.
2. For external dependencies, do not reference the version directly. Instead, use the version catalog feature.

Handling Null Return Values

1. In certain situations, null may need to be returned from a method, passed as a parameter, or set on a field. Only use Optional if a method is part of a fluent API. Since the runtime will rarely require this, the project standard is to use the org.jetbrains.annotations.Nullable and org.jetbrains.annotations.NotNull annotations.

Objects Serialization/Deserialization

1. TypeManager is the component responsible for json ser/des, you can also use the ObjectMapper inside it, but there should be no other ObjectMapper instance.

Class Naming

1. A single implementor of an interface should be named <interface name>Impl.
2. An implementor who is meant to be the default implementation for an interface but other are/can be defined used instead.

Observability

Services can implement one the following interfaces: ReadinessProvider, LivenessProvider, StartupStatusProvider and register with the HealthCheckService, thus reporting the service's readiness status through the observability API.

Streams

1. Always close explicitly Stream objects that are returned by a service/store, since they could carry a connection, which otherwise would leak.