Added javadoc to CommandDispatcher
This commit is contained in:
parent
1fe283c136
commit
8e9859e471
1 changed files with 284 additions and 0 deletions
|
@ -29,9 +29,27 @@ import java.util.concurrent.CompletableFuture;
|
||||||
import java.util.function.Predicate;
|
import java.util.function.Predicate;
|
||||||
import java.util.stream.Collectors;
|
import java.util.stream.Collectors;
|
||||||
|
|
||||||
|
|
||||||
|
/**
|
||||||
|
* The core command dispatcher, for registering, parsing, and executing commands.
|
||||||
|
*
|
||||||
|
* @param <S> a custom "source" type, such as a user or originator of a command
|
||||||
|
*/
|
||||||
public class CommandDispatcher<S> {
|
public class CommandDispatcher<S> {
|
||||||
|
/**
|
||||||
|
* The string required to separate individual arguments in an input string
|
||||||
|
*
|
||||||
|
* @see #ARGUMENT_SEPARATOR_CHAR
|
||||||
|
*/
|
||||||
public static final String ARGUMENT_SEPARATOR = " ";
|
public static final String ARGUMENT_SEPARATOR = " ";
|
||||||
|
|
||||||
|
/**
|
||||||
|
* The char required to separate individual arguments in an input string
|
||||||
|
*
|
||||||
|
* @see #ARGUMENT_SEPARATOR
|
||||||
|
*/
|
||||||
public static final char ARGUMENT_SEPARATOR_CHAR = ' ';
|
public static final char ARGUMENT_SEPARATOR_CHAR = ' ';
|
||||||
|
|
||||||
private static final String USAGE_OPTIONAL_OPEN = "[";
|
private static final String USAGE_OPTIONAL_OPEN = "[";
|
||||||
private static final String USAGE_OPTIONAL_CLOSE = "]";
|
private static final String USAGE_OPTIONAL_CLOSE = "]";
|
||||||
private static final String USAGE_REQUIRED_OPEN = "(";
|
private static final String USAGE_REQUIRED_OPEN = "(";
|
||||||
|
@ -48,33 +66,144 @@ public class CommandDispatcher<S> {
|
||||||
private ResultConsumer<S> consumer = (c, s, r) -> {
|
private ResultConsumer<S> consumer = (c, s, r) -> {
|
||||||
};
|
};
|
||||||
|
|
||||||
|
/**
|
||||||
|
* Create a new {@link CommandDispatcher} with the specified root node.
|
||||||
|
*
|
||||||
|
* <p>This is often useful to copy existing or pre-defined command trees.</p>
|
||||||
|
*
|
||||||
|
* @param root the existing {@link RootCommandNode} to use as the basis for this tree
|
||||||
|
*/
|
||||||
public CommandDispatcher(final RootCommandNode<S> root) {
|
public CommandDispatcher(final RootCommandNode<S> root) {
|
||||||
this.root = root;
|
this.root = root;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
/**
|
||||||
|
* Creates a new {@link CommandDispatcher} with an empty command tree.
|
||||||
|
*/
|
||||||
public CommandDispatcher() {
|
public CommandDispatcher() {
|
||||||
this(new RootCommandNode<>());
|
this(new RootCommandNode<>());
|
||||||
}
|
}
|
||||||
|
|
||||||
|
/**
|
||||||
|
* Utility method for registering new commands.
|
||||||
|
*
|
||||||
|
* <p>This is a shortcut for calling {@link RootCommandNode#addChild(CommandNode)} after building the provided {@code command}.</p>
|
||||||
|
*
|
||||||
|
* <p>As {@link RootCommandNode} can only hold literals, this method will only allow literal arguments.</p>
|
||||||
|
*
|
||||||
|
* @param command a literal argument builder to add to this command tree
|
||||||
|
* @return the node added to this tree
|
||||||
|
*/
|
||||||
public LiteralCommandNode<S> register(final LiteralArgumentBuilder<S> command) {
|
public LiteralCommandNode<S> register(final LiteralArgumentBuilder<S> command) {
|
||||||
final LiteralCommandNode<S> build = command.build();
|
final LiteralCommandNode<S> build = command.build();
|
||||||
root.addChild(build);
|
root.addChild(build);
|
||||||
return build;
|
return build;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
/**
|
||||||
|
* Sets a callback to be informed of the result of every command.
|
||||||
|
*
|
||||||
|
* @param consumer the new result consumer to be called
|
||||||
|
*/
|
||||||
public void setConsumer(final ResultConsumer<S> consumer) {
|
public void setConsumer(final ResultConsumer<S> consumer) {
|
||||||
this.consumer = consumer;
|
this.consumer = consumer;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
/**
|
||||||
|
* Parses and executes a given command.
|
||||||
|
*
|
||||||
|
* <p>This is a shortcut to first {@link #parse(StringReader, Object)} and then {@link #execute(ParseResults)}.</p>
|
||||||
|
*
|
||||||
|
* <p>It is recommended to parse and execute as separate steps, as parsing is often the most expensive step, and easiest to cache.</p>
|
||||||
|
*
|
||||||
|
* <p>If this command returns a value, then it successfully executed something. If it could not parse the command, or the execution was a failure,
|
||||||
|
* then an exception will be thrown. Most exceptions will be of type {@link CommandSyntaxException}, but it is possible that a {@link RuntimeException}
|
||||||
|
* may bubble up from the result of a command. The meaning behind the returned result is arbitrary, and will depend
|
||||||
|
* entirely on what command was performed.</p>
|
||||||
|
*
|
||||||
|
* <p>If the command passes through a node that is {@link CommandNode#isFork()} then it will be 'forked'.
|
||||||
|
* A forked command will not bubble up any {@link CommandSyntaxException}s, and the 'result' returned will turn into
|
||||||
|
* 'amount of successful commands executes'.</p>
|
||||||
|
*
|
||||||
|
* <p>After each and any command is ran, a registered callback given to {@link #setConsumer(ResultConsumer)}
|
||||||
|
* will be notified of the result and success of the command. You can use that method to gather more meaningful
|
||||||
|
* results than this method will return, especially when a command forks.</p>
|
||||||
|
*
|
||||||
|
* @param input a command string to parse & execute
|
||||||
|
* @param source a custom "source" object, usually representing the originator of this command
|
||||||
|
* @return a numeric result from a "command" that was performed
|
||||||
|
* @throws CommandSyntaxException if the command failed to parse or execute
|
||||||
|
* @throws RuntimeException if the command failed to execute and was not handled gracefully
|
||||||
|
* @see #parse(String, Object)
|
||||||
|
* @see #parse(StringReader, Object)
|
||||||
|
* @see #execute(ParseResults)
|
||||||
|
* @see #execute(StringReader, Object)
|
||||||
|
*/
|
||||||
public int execute(final String input, final S source) throws CommandSyntaxException {
|
public int execute(final String input, final S source) throws CommandSyntaxException {
|
||||||
return execute(new StringReader(input), source);
|
return execute(new StringReader(input), source);
|
||||||
}
|
}
|
||||||
|
|
||||||
|
/**
|
||||||
|
* Parses and executes a given command.
|
||||||
|
*
|
||||||
|
* <p>This is a shortcut to first {@link #parse(StringReader, Object)} and then {@link #execute(ParseResults)}.</p>
|
||||||
|
*
|
||||||
|
* <p>It is recommended to parse and execute as separate steps, as parsing is often the most expensive step, and easiest to cache.</p>
|
||||||
|
*
|
||||||
|
* <p>If this command returns a value, then it successfully executed something. If it could not parse the command, or the execution was a failure,
|
||||||
|
* then an exception will be thrown. Most exceptions will be of type {@link CommandSyntaxException}, but it is possible that a {@link RuntimeException}
|
||||||
|
* may bubble up from the result of a command. The meaning behind the returned result is arbitrary, and will depend
|
||||||
|
* entirely on what command was performed.</p>
|
||||||
|
*
|
||||||
|
* <p>If the command passes through a node that is {@link CommandNode#isFork()} then it will be 'forked'.
|
||||||
|
* A forked command will not bubble up any {@link CommandSyntaxException}s, and the 'result' returned will turn into
|
||||||
|
* 'amount of successful commands executes'.</p>
|
||||||
|
*
|
||||||
|
* <p>After each and any command is ran, a registered callback given to {@link #setConsumer(ResultConsumer)}
|
||||||
|
* will be notified of the result and success of the command. You can use that method to gather more meaningful
|
||||||
|
* results than this method will return, especially when a command forks.</p>
|
||||||
|
*
|
||||||
|
* @param input a command string to parse & execute
|
||||||
|
* @param source a custom "source" object, usually representing the originator of this command
|
||||||
|
* @return a numeric result from a "command" that was performed
|
||||||
|
* @throws CommandSyntaxException if the command failed to parse or execute
|
||||||
|
* @throws RuntimeException if the command failed to execute and was not handled gracefully
|
||||||
|
* @see #parse(String, Object)
|
||||||
|
* @see #parse(StringReader, Object)
|
||||||
|
* @see #execute(ParseResults)
|
||||||
|
* @see #execute(String, Object)
|
||||||
|
*/
|
||||||
public int execute(final StringReader input, final S source) throws CommandSyntaxException {
|
public int execute(final StringReader input, final S source) throws CommandSyntaxException {
|
||||||
final ParseResults<S> parse = parse(input, source);
|
final ParseResults<S> parse = parse(input, source);
|
||||||
return execute(parse);
|
return execute(parse);
|
||||||
}
|
}
|
||||||
|
|
||||||
|
/**
|
||||||
|
* Executes a given pre-parsed command.
|
||||||
|
*
|
||||||
|
* <p>If this command returns a value, then it successfully executed something. If the execution was a failure,
|
||||||
|
* then an exception will be thrown.
|
||||||
|
* Most exceptions will be of type {@link CommandSyntaxException}, but it is possible that a {@link RuntimeException}
|
||||||
|
* may bubble up from the result of a command. The meaning behind the returned result is arbitrary, and will depend
|
||||||
|
* entirely on what command was performed.</p>
|
||||||
|
*
|
||||||
|
* <p>If the command passes through a node that is {@link CommandNode#isFork()} then it will be 'forked'.
|
||||||
|
* A forked command will not bubble up any {@link CommandSyntaxException}s, and the 'result' returned will turn into
|
||||||
|
* 'amount of successful commands executes'.</p>
|
||||||
|
*
|
||||||
|
* <p>After each and any command is ran, a registered callback given to {@link #setConsumer(ResultConsumer)}
|
||||||
|
* will be notified of the result and success of the command. You can use that method to gather more meaningful
|
||||||
|
* results than this method will return, especially when a command forks.</p>
|
||||||
|
*
|
||||||
|
* @param parse the result of a successful {@link #parse(StringReader, Object)}
|
||||||
|
* @return a numeric result from a "command" that was performed.
|
||||||
|
* @throws CommandSyntaxException if the command failed to parse or execute
|
||||||
|
* @throws RuntimeException if the command failed to execute and was not handled gracefully
|
||||||
|
* @see #parse(String, Object)
|
||||||
|
* @see #parse(StringReader, Object)
|
||||||
|
* @see #execute(String, Object)
|
||||||
|
* @see #execute(StringReader, Object)
|
||||||
|
*/
|
||||||
public int execute(final ParseResults<S> parse) throws CommandSyntaxException {
|
public int execute(final ParseResults<S> parse) throws CommandSyntaxException {
|
||||||
if (parse.getReader().canRead()) {
|
if (parse.getReader().canRead()) {
|
||||||
if (parse.getExceptions().size() == 1) {
|
if (parse.getExceptions().size() == 1) {
|
||||||
|
@ -157,10 +286,64 @@ public class CommandDispatcher<S> {
|
||||||
return forked ? successfulForks : result;
|
return forked ? successfulForks : result;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
/**
|
||||||
|
* Parses a given command.
|
||||||
|
*
|
||||||
|
* <p>The result of this method can be cached, and it is advised to do so where appropriate. Parsing is often the
|
||||||
|
* most expensive step, and this allows you to essentially "precompile" a command if it will be ran often.</p>
|
||||||
|
*
|
||||||
|
* <p>If the command passes through a node that is {@link CommandNode#isFork()} then the resulting context will be marked as 'forked'.
|
||||||
|
* Forked contexts may contain child contexts, which may be modified by the {@link RedirectModifier} attached to the fork.</p>
|
||||||
|
*
|
||||||
|
* <p>Parsing a command can never fail, you will always be provided with a new {@link ParseResults}.
|
||||||
|
* However, that does not mean that it will always parse into a valid command. You should inspect the returned results
|
||||||
|
* to check for validity. If its {@link ParseResults#getReader()} {@link StringReader#canRead()} then it did not finish
|
||||||
|
* parsing successfully. You can use that position as an indicator to the user where the command stopped being valid.
|
||||||
|
* You may inspect {@link ParseResults#getExceptions()} if you know the parse failed, as it will explain why it could
|
||||||
|
* not find any valid commands. It may contain multiple exceptions, one for each "potential node" that it could have visited,
|
||||||
|
* explaining why it did not go down that node.</p>
|
||||||
|
*
|
||||||
|
* <p>When you eventually call {@link #execute(ParseResults)} with the result of this method, the above error checking
|
||||||
|
* will occur. You only need to inspect it yourself if you wish to handle that yourself.</p>
|
||||||
|
*
|
||||||
|
* @param command a command string to parse
|
||||||
|
* @param source a custom "source" object, usually representing the originator of this command
|
||||||
|
* @return the result of parsing this command
|
||||||
|
* @see #parse(StringReader, Object)
|
||||||
|
* @see #execute(ParseResults)
|
||||||
|
* @see #execute(String, Object)
|
||||||
|
*/
|
||||||
public ParseResults<S> parse(final String command, final S source) {
|
public ParseResults<S> parse(final String command, final S source) {
|
||||||
return parse(new StringReader(command), source);
|
return parse(new StringReader(command), source);
|
||||||
}
|
}
|
||||||
|
|
||||||
|
/**
|
||||||
|
* Parses a given command.
|
||||||
|
*
|
||||||
|
* <p>The result of this method can be cached, and it is advised to do so where appropriate. Parsing is often the
|
||||||
|
* most expensive step, and this allows you to essentially "precompile" a command if it will be ran often.</p>
|
||||||
|
*
|
||||||
|
* <p>If the command passes through a node that is {@link CommandNode#isFork()} then the resulting context will be marked as 'forked'.
|
||||||
|
* Forked contexts may contain child contexts, which may be modified by the {@link RedirectModifier} attached to the fork.</p>
|
||||||
|
*
|
||||||
|
* <p>Parsing a command can never fail, you will always be provided with a new {@link ParseResults}.
|
||||||
|
* However, that does not mean that it will always parse into a valid command. You should inspect the returned results
|
||||||
|
* to check for validity. If its {@link ParseResults#getReader()} {@link StringReader#canRead()} then it did not finish
|
||||||
|
* parsing successfully. You can use that position as an indicator to the user where the command stopped being valid.
|
||||||
|
* You may inspect {@link ParseResults#getExceptions()} if you know the parse failed, as it will explain why it could
|
||||||
|
* not find any valid commands. It may contain multiple exceptions, one for each "potential node" that it could have visited,
|
||||||
|
* explaining why it did not go down that node.</p>
|
||||||
|
*
|
||||||
|
* <p>When you eventually call {@link #execute(ParseResults)} with the result of this method, the above error checking
|
||||||
|
* will occur. You only need to inspect it yourself if you wish to handle that yourself.</p>
|
||||||
|
*
|
||||||
|
* @param command a command string to parse
|
||||||
|
* @param source a custom "source" object, usually representing the originator of this command
|
||||||
|
* @return the result of parsing this command
|
||||||
|
* @see #parse(String, Object)
|
||||||
|
* @see #execute(ParseResults)
|
||||||
|
* @see #execute(String, Object)
|
||||||
|
*/
|
||||||
public ParseResults<S> parse(final StringReader command, final S source) {
|
public ParseResults<S> parse(final StringReader command, final S source) {
|
||||||
final CommandContextBuilder<S> context = new CommandContextBuilder<>(this, source, 0);
|
final CommandContextBuilder<S> context = new CommandContextBuilder<>(this, source, 0);
|
||||||
return parseNodes(root, command, context);
|
return parseNodes(root, command, context);
|
||||||
|
@ -256,6 +439,27 @@ public class CommandDispatcher<S> {
|
||||||
return new ParseResults<>(contextSoFar, originalReader.getCursor(), originalReader, errors == null ? Collections.emptyMap() : errors);
|
return new ParseResults<>(contextSoFar, originalReader.getCursor(), originalReader, errors == null ? Collections.emptyMap() : errors);
|
||||||
}
|
}
|
||||||
|
|
||||||
|
/**
|
||||||
|
* Gets all possible executable commands following the given node.
|
||||||
|
*
|
||||||
|
* <p>You may use {@link #getRoot()} as a target to get all usage data for the entire command tree.</p>
|
||||||
|
*
|
||||||
|
* <p>The returned syntax will be in "simple" form: {@code <param>} and {@code literal}. "Optional" nodes will be
|
||||||
|
* listed as multiple entries: the parent node, and the child nodes.
|
||||||
|
* For example, a required literal "foo" followed by an optional param "int" will be two nodes:</p>
|
||||||
|
* <ul>
|
||||||
|
* <li>{@code foo}</li>
|
||||||
|
* <li>{@code foo <int>}</li>
|
||||||
|
* </ul>
|
||||||
|
*
|
||||||
|
* <p>The path to the specified node will <b>not</b> be prepended to the output, as there can theoretically be many
|
||||||
|
* ways to reach a given node. It will only give you paths relative to the specified node, not absolute from root.</p>
|
||||||
|
*
|
||||||
|
* @param node target node to get child usage strings for
|
||||||
|
* @param source a custom "source" object, usually representing the originator of this command
|
||||||
|
* @param restricted if true, commands that the {@code source} cannot access will not be mentioned
|
||||||
|
* @return array of full usage strings under the target node
|
||||||
|
*/
|
||||||
public String[] getAllUsage(final CommandNode<S> node, final S source, final boolean restricted) {
|
public String[] getAllUsage(final CommandNode<S> node, final S source, final boolean restricted) {
|
||||||
final ArrayList<String> result = Lists.newArrayList();
|
final ArrayList<String> result = Lists.newArrayList();
|
||||||
getAllUsage(node, source, result, "", restricted);
|
getAllUsage(node, source, result, "", restricted);
|
||||||
|
@ -281,6 +485,27 @@ public class CommandDispatcher<S> {
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
/**
|
||||||
|
* Gets the possible executable commands from a specified node.
|
||||||
|
*
|
||||||
|
* <p>You may use {@link #getRoot()} as a target to get usage data for the entire command tree.</p>
|
||||||
|
*
|
||||||
|
* <p>The returned syntax will be in "smart" form: {@code <param>}, {@code literal}, {@code [optional]} and {@code (either|or)}.
|
||||||
|
* These forms may be mixed and matched to provide as much information about the child nodes as it can, without being too verbose.
|
||||||
|
* For example, a required literal "foo" followed by an optional param "int" can be compressed into one string:</p>
|
||||||
|
* <ul>
|
||||||
|
* <li>{@code foo [<int>]}</li>
|
||||||
|
* </ul>
|
||||||
|
*
|
||||||
|
* <p>The path to the specified node will <b>not</b> be prepended to the output, as there can theoretically be many
|
||||||
|
* ways to reach a given node. It will only give you paths relative to the specified node, not absolute from root.</p>
|
||||||
|
*
|
||||||
|
* <p>The returned usage will be restricted to only commands that the provided {@code source} can use.</p>
|
||||||
|
*
|
||||||
|
* @param node target node to get child usage strings for
|
||||||
|
* @param source a custom "source" object, usually representing the originator of this command
|
||||||
|
* @return array of full usage strings under the target node
|
||||||
|
*/
|
||||||
public Map<CommandNode<S>, String> getSmartUsage(final CommandNode<S> node, final S source) {
|
public Map<CommandNode<S>, String> getSmartUsage(final CommandNode<S> node, final S source) {
|
||||||
final Map<CommandNode<S>, String> result = Maps.newLinkedHashMap();
|
final Map<CommandNode<S>, String> result = Maps.newLinkedHashMap();
|
||||||
|
|
||||||
|
@ -348,6 +573,21 @@ public class CommandDispatcher<S> {
|
||||||
return self;
|
return self;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
/**
|
||||||
|
* Gets suggestions for a parsed input string on what comes next.
|
||||||
|
*
|
||||||
|
* <p>As it is ultimately up to custom argument types to provide suggestions, it may be an asynchronous operation,
|
||||||
|
* for example getting in-game data or player names etc. As such, this method returns a future and no guarantees
|
||||||
|
* are made to when or how the future completes.</p>
|
||||||
|
*
|
||||||
|
* <p>The suggestions provided will be in the context of the end of the parsed input string, but may suggest
|
||||||
|
* new or replacement strings for earlier in the input string. For example, if the end of the string was
|
||||||
|
* {@code foobar} but an argument preferred it to be {@code minecraft:foobar}, it will suggest a replacement for that
|
||||||
|
* whole segment of the input.</p>
|
||||||
|
*
|
||||||
|
* @param parse the result of a {@link #parse(StringReader, Object)}
|
||||||
|
* @return a future that will eventually resolve into a {@link Suggestions} object
|
||||||
|
*/
|
||||||
public CompletableFuture<Suggestions> getCompletionSuggestions(final ParseResults<S> parse) {
|
public CompletableFuture<Suggestions> getCompletionSuggestions(final ParseResults<S> parse) {
|
||||||
final CommandContextBuilder<S> rootContext = parse.getContext();
|
final CommandContextBuilder<S> rootContext = parse.getContext();
|
||||||
final CommandContextBuilder<S> context = rootContext.getLastChild();
|
final CommandContextBuilder<S> context = rootContext.getLastChild();
|
||||||
|
@ -397,10 +637,33 @@ public class CommandDispatcher<S> {
|
||||||
return result;
|
return result;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
/**
|
||||||
|
* Gets the root of this command tree.
|
||||||
|
*
|
||||||
|
* <p>This is often useful as a target of a {@link com.mojang.brigadier.builder.ArgumentBuilder#redirect(CommandNode)},
|
||||||
|
* {@link #getAllUsage(CommandNode, Object, boolean)} or {@link #getSmartUsage(CommandNode, Object)}.
|
||||||
|
* You may also use it to clone the command tree via {@link #CommandDispatcher(RootCommandNode)}.</p>
|
||||||
|
*
|
||||||
|
* @return root of the command tree
|
||||||
|
*/
|
||||||
public RootCommandNode<S> getRoot() {
|
public RootCommandNode<S> getRoot() {
|
||||||
return root;
|
return root;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
/**
|
||||||
|
* Finds a valid path to a given node on the command tree.
|
||||||
|
*
|
||||||
|
* <p>There may theoretically be multiple paths to a node on the tree, especially with the use of forking or redirecting.
|
||||||
|
* As such, this method makes no guarantees about which path it finds. It will not look at forks or redirects,
|
||||||
|
* and find the first instance of the target node on the tree.</p>
|
||||||
|
*
|
||||||
|
* <p>The only guarantee made is that for the same command tree and the same version of this library, the result of
|
||||||
|
* this method will <b>always</b> be a valid input for {@link #findNode(Collection)}, which should return the same node
|
||||||
|
* as provided to this method.</p>
|
||||||
|
*
|
||||||
|
* @param target the target node you are finding a path for
|
||||||
|
* @return a path to the resulting node, or an empty list if it was not found
|
||||||
|
*/
|
||||||
public Collection<String> getPath(final CommandNode<S> target) {
|
public Collection<String> getPath(final CommandNode<S> target) {
|
||||||
final List<List<CommandNode<S>>> nodes = new ArrayList<>();
|
final List<List<CommandNode<S>>> nodes = new ArrayList<>();
|
||||||
addPaths(root, nodes, new ArrayList<>());
|
addPaths(root, nodes, new ArrayList<>());
|
||||||
|
@ -420,6 +683,17 @@ public class CommandDispatcher<S> {
|
||||||
return Collections.emptyList();
|
return Collections.emptyList();
|
||||||
}
|
}
|
||||||
|
|
||||||
|
/**
|
||||||
|
* Finds a node by its path
|
||||||
|
*
|
||||||
|
* <p>Paths may be generated with {@link #getPath(CommandNode)}, and are guaranteed (for the same tree, and the
|
||||||
|
* same version of this library) to always produce the same valid node by this method.</p>
|
||||||
|
*
|
||||||
|
* <p>If a node could not be found at the specified path, then {@code null} will be returned.</p>
|
||||||
|
*
|
||||||
|
* @param path a generated path to a node
|
||||||
|
* @return the node at the given path, or null if not found
|
||||||
|
*/
|
||||||
public CommandNode<S> findNode(final Collection<String> path) {
|
public CommandNode<S> findNode(final Collection<String> path) {
|
||||||
CommandNode<S> node = root;
|
CommandNode<S> node = root;
|
||||||
for (final String name : path) {
|
for (final String name : path) {
|
||||||
|
@ -431,6 +705,16 @@ public class CommandDispatcher<S> {
|
||||||
return node;
|
return node;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
/**
|
||||||
|
* Scans the command tree for potential ambiguous commands.
|
||||||
|
*
|
||||||
|
* <p>This is a shortcut for {@link CommandNode#findAmbiguities(AmbiguityConsumer)} on {@link #getRoot()}.</p>
|
||||||
|
*
|
||||||
|
* <p>Ambiguities are detected by testing every {@link CommandNode#getExamples()} on one node verses every sibling
|
||||||
|
* node. This is not fool proof, and relies a lot on the providers of the used argument types to give good examples.</p>
|
||||||
|
*
|
||||||
|
* @param consumer a callback to be notified of potential ambiguities
|
||||||
|
*/
|
||||||
public void findAmbiguities(final AmbiguityConsumer<S> consumer) {
|
public void findAmbiguities(final AmbiguityConsumer<S> consumer) {
|
||||||
root.findAmbiguities(consumer);
|
root.findAmbiguities(consumer);
|
||||||
}
|
}
|
||||||
|
|
Loading…
Reference in a new issue