* Use your judgment always. Any rule can be broken with a good reason. Don't follow a rule without understanding its purpose.
* Write code for readability above all else. Always think about how another developer would work with your code.
* Avoid repetitive code. Factor out the repetition, if there is a reasonable way to do so.
* We have a lot of legacy code lying around. Be careful not to build on obsolete systems or use outdated conventions.
* Formatting
* For basic things like tab size and bracket placement, just match the existing code.
* Use whatever layout you feel makes the code most readable, even if that differs from case to case.
* You can use concise formats in places where it helps readability (e.g. single-line getters).
* Comments
* Try to write code that is obvious enough so that it doesn't need to be explained with comments.
* In places where a reader might be confused or miss something important, use comments to fill them in.
* Don't put redundant or obvious things in comments or javadocs.
* Ensure your IDE is not inserting any generated comments.
* Nulls
* Strongly prefer `java.util.Optional` over `null`, generally speaking.
* Use empty collections as nil values for collections.
* Use `@Nullable` wherever nulls are allowed. Place it before the type, if possible.
* Don't use `@Nonnull`. Assume anything (in our own code) without `@Nullable` is never null.
* Use `checkNotNull` on constructor arguments for manually created objects. Don't check `@Inject`ed values, they cannot be null.
* Structure
* Design classes to do [one thing only](https://en.wikipedia.org/wiki/Single_responsibility_principle).
If a class provides multiple services, break them down into seperate public interfaces and keep the class private.
* Use `final` fields, and create immutable data types, wherever possible.
* Don't create unnecessary getters and setters, only what is actually used.
* No mutable static fields, collections, or any other static state (there are a few exceptions, such as caches and `ThreadLocal`s).
* Getters don't have to start with `get`, but they can if you think it's important.
* Injection
* We use Guice everywhere. You will need to [understand it thoroughly](https://github.com/google/guice/wiki/Motivation).
* Follow the Guice [best practices](https://github.com/google/guice/wiki/InjectOnlyDirectDependencies) (especially that one).
* We sometimes use the term "manifest" in place of "module", to avoid confusion with PGM modules.
* Using Guice in a Bukkit plugin environment has proven to be [somewhat complex](https://github.com/OvercastNetwork/Plugins/blob/master/Util/core/src/main/java/tc/oc/commons/core/plugin/InjectedPluginLoader.java).
You don't have to understand all of that, but there are a few things you should know:
* Each plugin has its own [private module](https://google.github.io/guice/api-docs/latest/javadoc/index.html?com/google/inject/PrivateModule.html)
in which most of its bindings live.
* Each plugin instance is bound to `org.bukkit.plugin.Plugin` inside its private module.
* Anything that indirectly depends on `Plugin` will need to be bound in some plugin's private module.
* If other plugins need access to that thing, it needs to be [exposed](https://google.github.io/guice/api-docs/latest/javadoc/com/google/inject/PrivateModule.html#expose-com.google.inject.Key-).
* Avoid depending on `Plugin` directly. There are specific bindings for most Bukkit service types (e.g. `Configuration`)
and several interfaces of our own that wrap small parts of the Bukkit API (e.g. `BukkitEventBus`).
* If you really need a `Plugin`, always inject the base interface, never a specific plugin's class. This makes it easy to move things between plugins.
* If the same `@Singleton` type is provisioned in multiple plugins, we will detect it and throw an exception.
If you actually want a per-plugin singleton, make it `@PluginScoped`.
*`PluginFacet`s are service objects registered with a specific plugin to share its lifecycle callbacks (i.e. enable() and disable()).
They are also registered automatically in the appropriate way if they implement various interfaces such as `Listener` (see the javadocs for details).
* Exceptions
* Detect errors as early as possible, ideally at server startup. This applies to both user errors and internal assertions.
* Only catch specific exceptions that you are expecting and can handle thoroughly. Don't hide exceptions that other handlers need to know about.
* Avoid catching common exceptions like `IllegalArgumentException`, because it's hard to be certain where they come from.
If you need to catch them, keep the code inside the `try` block as small as possible.
* Don't catch all exceptions or try to handle internal errors for no particular reason. We have a nice framework for dealing with unhandled exceptions at the top-level.
* If you need to implement a general exception handler, inject a `tc.oc.commons.core.exception.ExceptionHandler` and pass it the exceptions.
* Ensure that all errors are seen by a human who knows how to fix them:
* Pure user errors only need to be sent to the user who provided the bad input
* Map errors can be sent to builders in-game (and in Sentry too, on production servers) using the `MapdevLogger`.
* Any unexpected exception on any server must notify developers somehow i.e. through the `LoggingExceptionHandler` or some system logger.
You can tell the user about the error so they know what's going on, but don't expect them to deal with it.
* Concurrency
* Avoid concurrency. It's hard, and we don't have a general solution for doing it safely.
* A Bukkit server has a single "main thread" where most game logic runs, and multiple background threads for I/O and serialization.
All `Event`s (except `AsyncEvent`s) and scheduled tasks (except async tasks) run on the main thread.
* Our own plugins use background thread pools for API operations, which use HTTP and AMQP.
* Never block the main thread on API calls, or any other I/O operation. Use `ListenableFuture`s and `FutureCallback`s to handle API results.
* Don't use the Bukkit API, or any of our own APIs, from a background thread, unless it is explicitly allowed by the API.
Use `MainThreadExecutor` or `SyncExecutor` to get back on the main thread from a background thread.
* A Bungee server is entirely multi-threaded. Handlers for a specific event run in sequence, but seperate events and tasks can run concurrently.
This is one of the reasons we avoid doing things in Bungee.
* Localization
* All in-game text must be localized.
* To add a text string to the game, put the english template in one of the `.properties` files in Commons.
* Use a `TranslatableComponent` to display the localized message. However, it must pass through the `ComponentRenderContext` before being
sent to the player. That will apply the server-side translations.
* Any changes to the templates must be sent to Crowdin before volunteers can start translating them.
This is done with the [Crowdin integration tool](https://crowdin.com/page/cli-tool) by running `crowdin-cli upload sources` in the Commons folder.
This should be done after the new code is deployed to production.
* Libraries
* TODO
* Utilities
* TODO
* Testing
* TODO
* Logging
* TODO
## Workflow
* We use Git, with a typical [feature branch workflow](https://www.atlassian.com/git/tutorials/comparing-workflows/feature-branch-workflow)
* Trivial changes and emergency fixes can be merged straight to the master branch
* Any significant change requires a PR, and code review by at least one other developer.
This applies indiscriminately to all developers. Everyone should have their code reviewed, and anyone can review anyone else's code.
* Once a change has been merged to master, it should be deployed ASAP so that problems can be found.
Deploying several old changes at once just makes it harder to trace bugs to their source.
* Without automated tests, we rely heavily on user reports and Sentry alerts to discover regressions.
Developers should be around for at least a few hours after their change is deployed, in case something breaks.
