Added readme
This commit is contained in:
parent
189c7d0433
commit
1fe283c136
1 changed files with 133 additions and 0 deletions
133
README.md
Normal file
133
README.md
Normal file
|
@ -0,0 +1,133 @@
|
||||||
|
# Brigadier
|
||||||
|
Brigadier is a command parser & dispatcher, designed and developed for Minecraft: Java Edition and now freely available for use elsewhere under the MIT license.
|
||||||
|
|
||||||
|
# Installation
|
||||||
|
Brigadier is available to Maven & Gradle via `libraries.minecraft.net`. Its group is `com.mojang`, and artifact name is `brigadier`.
|
||||||
|
|
||||||
|
## Gradle
|
||||||
|
First include our repository:
|
||||||
|
```groovy
|
||||||
|
maven {
|
||||||
|
url "https://libraries.minecraft.net"
|
||||||
|
}
|
||||||
|
```
|
||||||
|
|
||||||
|
And then use this library (change `(the latest version)` to the latest version!):
|
||||||
|
```groovy
|
||||||
|
compile 'com.mojang:brigadier:(the latest version)'
|
||||||
|
```
|
||||||
|
|
||||||
|
## Maven
|
||||||
|
First include our repository:
|
||||||
|
```xml
|
||||||
|
<repository>
|
||||||
|
<id>minecraft-libraries</id>
|
||||||
|
<name>Minecraft Libraries</name>
|
||||||
|
<url>https://libraries.minecraft.net</url>
|
||||||
|
</repository>
|
||||||
|
```
|
||||||
|
|
||||||
|
And then use this library (change `(the latest version)` to the latest version!):
|
||||||
|
```xml
|
||||||
|
<dependency>
|
||||||
|
<groupId>com.mojang</groupId>
|
||||||
|
<artifactId>brigadier</artifactId>
|
||||||
|
<version>(the latest version)</version>
|
||||||
|
</dependency>
|
||||||
|
```
|
||||||
|
|
||||||
|
# Contributing
|
||||||
|
Contributions are suspended until we have a process in place to handle them.
|
||||||
|
|
||||||
|
Most contributions will require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to,
|
||||||
|
and actually do, grant us the rights to use your contribution. For details, visit https://cla.microsoft.com.
|
||||||
|
|
||||||
|
This project has adopted the [Microsoft Open Source Code of Conduct](https://opensource.microsoft.com/codeofconduct/).
|
||||||
|
For more information see the [Code of Conduct FAQ](https://opensource.microsoft.com/codeofconduct/faq/) or
|
||||||
|
contact [opencode@microsoft.com](mailto:opencode@microsoft.com) with any additional questions or comments.
|
||||||
|
|
||||||
|
# Usage
|
||||||
|
At the heart of Brigadier, you need a `CommandDispatcher<S>`, where `<S>` is any custom object you choose to identify a "command source".
|
||||||
|
|
||||||
|
A command dispatcher holds a "command tree", which are a series of `CommandNode` which represent the various possible syntax that form a valid command.
|
||||||
|
|
||||||
|
## Registering a new command
|
||||||
|
Before we can start parsing and dispatching commands, we need to build up our command tree. Every registration is an append operation,
|
||||||
|
so you can freely extend existing commands in a project without needing access to the source code that created them.
|
||||||
|
|
||||||
|
Command registration also encourages use of a builder pattern to keep code cruft to a minimum.
|
||||||
|
|
||||||
|
A "command" is a fairly loose term, but typically it means an exit point of the command tree.
|
||||||
|
Every node can have an `executes` function attached to it, which signifies that if the input stops here then this function will be called with the context so far.
|
||||||
|
|
||||||
|
Consider the following example:
|
||||||
|
```java
|
||||||
|
CommandDispatcher<CommandSourceStack> dispatcher = new CommandDispatcher();
|
||||||
|
|
||||||
|
dispatcher.register(
|
||||||
|
literal("foo")
|
||||||
|
.then(
|
||||||
|
argument("bar", integer())
|
||||||
|
.executes(c -> {
|
||||||
|
System.out.println("Bar is " + getInteger(c, "bar"));
|
||||||
|
return 1;
|
||||||
|
})
|
||||||
|
)
|
||||||
|
.executes(c -> {
|
||||||
|
System.out.println("Called foo with no arguments");
|
||||||
|
return 1;
|
||||||
|
})
|
||||||
|
);
|
||||||
|
```
|
||||||
|
|
||||||
|
This snippet registers two "commands": `foo` and `foo <bar>`. It is also common to refer to the `<bar>` as a "subcommand" of `foo`, as it's a child node.
|
||||||
|
|
||||||
|
At the start of the tree is a "root node", and it **must** have `LiteralCommandNode`s as children. Here, we register one command under the root: `literal("foo")`, which means "the user must type the literal string 'foo'".
|
||||||
|
|
||||||
|
Under that is two extra definitions: a child node for possible further evaluation, or an `executes` block if the user input stops here.
|
||||||
|
|
||||||
|
The child node works exactly the same way, but is no longer limited to literals. The other type of node that is now allowed is an `ArgumentCommandNode`, which takes in a name and an argument type.
|
||||||
|
|
||||||
|
Arguments can be anything, and you are encouraged to build your own for seamless integration into your own product. There are some standard arguments included in brigadier, such as `IntegerArgumentType`.
|
||||||
|
|
||||||
|
Argument types will be asked to parse input as much as they can, and then store the "result" of that argument however they see fit or throw a relevant error if they can't parse.
|
||||||
|
|
||||||
|
For example, an integer argument would parse "123" and store it as `123` (`int`), but throw an error if the input were `onetwothree`.
|
||||||
|
|
||||||
|
When a command is actually ran, it can access these arguments in the context provided to the registered function.
|
||||||
|
|
||||||
|
## Parsing user input
|
||||||
|
So, we've registered some commands and now we're ready to take in user input. If you're in a rush, you can just call `dispatcher.execute("foo 123", source)` and call it a day.
|
||||||
|
|
||||||
|
The result of `execute` is an integer returned by the command it evaluated. Its meaning varies depending on command, and typically will not be useful to programmers.
|
||||||
|
|
||||||
|
The `source` is an object of `<S>`, your own custom class to track users/players/etc. It will be provided to the command so that it has some context on what's happening.
|
||||||
|
|
||||||
|
If the command failed or could not parse, some form of `CommandSyntaxException` will be thrown. It is also possible for a `RuntimeException` to be bubbled up, if not properly handled in a command.
|
||||||
|
|
||||||
|
If you wish to have more control over the parsing & executing of commands, or wish to cache the parse results so you can execute it multiple times, you can split it up into two steps:
|
||||||
|
|
||||||
|
```java
|
||||||
|
final ParseResults<S> parse = dispatcher.parse("foo 123", source);
|
||||||
|
final int result = execute(parse);
|
||||||
|
```
|
||||||
|
|
||||||
|
This is highly recommended as the parse step is the most expensive, and may be easily cached depending on your application.
|
||||||
|
|
||||||
|
You can also use this to do further introspection on a command, before (or without) actually running it.
|
||||||
|
|
||||||
|
## Inspecting a command
|
||||||
|
If you `parse` some input, you can find out what it will perform (if anything) and provide hints to the user safely and immediately.
|
||||||
|
|
||||||
|
The parse will never fail, and the `ParseResults<S>` it returns will contain a *possible* context that a command may be called with
|
||||||
|
(and from that, you can inspect which nodes the user entered, complete with start/end positions in the input string).
|
||||||
|
It also contains a map of parse exceptions for each command node it encountered. If it couldn't build a valid context, then
|
||||||
|
the reason why is inside this exception map.
|
||||||
|
|
||||||
|
## Displaying usage info
|
||||||
|
There's two forms of "usage strings" provided by this library, both require a target node.
|
||||||
|
|
||||||
|
`getAllUsage(node, source, restricted)` will return a list of all possible commands (executable end-points) under the target node and their human readable path. If `restricted`, it will ignore commands that `source` does not have access to. This will look like [`foo`, `foo <bar>`]
|
||||||
|
|
||||||
|
`getSmartUsage(node, source)` will return a map of the child nodes to their "smart usage" human readable path. This tries to squash future-nodes together and show optional & typed information, and can look like `foo (<bar>)`
|
||||||
|
|
Loading…
Reference in a new issue