mirror of
https://codeberg.org/forgejo/forgejo.git
synced 2024-12-16 19:32:05 +03:00
69dae1ec1c
* Added coding style modes to SimpleMDE * Moved the CodeMirror addon from simplemd to codemirror directory
220 lines
7.4 KiB
JavaScript
220 lines
7.4 KiB
JavaScript
// CodeMirror, copyright (c) by Marijn Haverbeke and others
|
|
// Distributed under an MIT license: http://codemirror.net/LICENSE
|
|
|
|
(function(mod) {
|
|
if (typeof exports == "object" && typeof module == "object") // CommonJS
|
|
mod(require("../../lib/codemirror"));
|
|
else if (typeof define == "function" && define.amd) // AMD
|
|
define(["../../lib/codemirror"], mod);
|
|
else // Plain browser env
|
|
mod(CodeMirror);
|
|
})(function(CodeMirror) {
|
|
"use strict";
|
|
|
|
CodeMirror.defineMode("puppet", function () {
|
|
// Stores the words from the define method
|
|
var words = {};
|
|
// Taken, mostly, from the Puppet official variable standards regex
|
|
var variable_regex = /({)?([a-z][a-z0-9_]*)?((::[a-z][a-z0-9_]*)*::)?[a-zA-Z0-9_]+(})?/;
|
|
|
|
// Takes a string of words separated by spaces and adds them as
|
|
// keys with the value of the first argument 'style'
|
|
function define(style, string) {
|
|
var split = string.split(' ');
|
|
for (var i = 0; i < split.length; i++) {
|
|
words[split[i]] = style;
|
|
}
|
|
}
|
|
|
|
// Takes commonly known puppet types/words and classifies them to a style
|
|
define('keyword', 'class define site node include import inherits');
|
|
define('keyword', 'case if else in and elsif default or');
|
|
define('atom', 'false true running present absent file directory undef');
|
|
define('builtin', 'action augeas burst chain computer cron destination dport exec ' +
|
|
'file filebucket group host icmp iniface interface jump k5login limit log_level ' +
|
|
'log_prefix macauthorization mailalias maillist mcx mount nagios_command ' +
|
|
'nagios_contact nagios_contactgroup nagios_host nagios_hostdependency ' +
|
|
'nagios_hostescalation nagios_hostextinfo nagios_hostgroup nagios_service ' +
|
|
'nagios_servicedependency nagios_serviceescalation nagios_serviceextinfo ' +
|
|
'nagios_servicegroup nagios_timeperiod name notify outiface package proto reject ' +
|
|
'resources router schedule scheduled_task selboolean selmodule service source ' +
|
|
'sport ssh_authorized_key sshkey stage state table tidy todest toports tosource ' +
|
|
'user vlan yumrepo zfs zone zpool');
|
|
|
|
// After finding a start of a string ('|") this function attempts to find the end;
|
|
// If a variable is encountered along the way, we display it differently when it
|
|
// is encapsulated in a double-quoted string.
|
|
function tokenString(stream, state) {
|
|
var current, prev, found_var = false;
|
|
while (!stream.eol() && (current = stream.next()) != state.pending) {
|
|
if (current === '$' && prev != '\\' && state.pending == '"') {
|
|
found_var = true;
|
|
break;
|
|
}
|
|
prev = current;
|
|
}
|
|
if (found_var) {
|
|
stream.backUp(1);
|
|
}
|
|
if (current == state.pending) {
|
|
state.continueString = false;
|
|
} else {
|
|
state.continueString = true;
|
|
}
|
|
return "string";
|
|
}
|
|
|
|
// Main function
|
|
function tokenize(stream, state) {
|
|
// Matches one whole word
|
|
var word = stream.match(/[\w]+/, false);
|
|
// Matches attributes (i.e. ensure => present ; 'ensure' would be matched)
|
|
var attribute = stream.match(/(\s+)?\w+\s+=>.*/, false);
|
|
// Matches non-builtin resource declarations
|
|
// (i.e. "apache::vhost {" or "mycustomclasss {" would be matched)
|
|
var resource = stream.match(/(\s+)?[\w:_]+(\s+)?{/, false);
|
|
// Matches virtual and exported resources (i.e. @@user { ; and the like)
|
|
var special_resource = stream.match(/(\s+)?[@]{1,2}[\w:_]+(\s+)?{/, false);
|
|
|
|
// Finally advance the stream
|
|
var ch = stream.next();
|
|
|
|
// Have we found a variable?
|
|
if (ch === '$') {
|
|
if (stream.match(variable_regex)) {
|
|
// If so, and its in a string, assign it a different color
|
|
return state.continueString ? 'variable-2' : 'variable';
|
|
}
|
|
// Otherwise return an invalid variable
|
|
return "error";
|
|
}
|
|
// Should we still be looking for the end of a string?
|
|
if (state.continueString) {
|
|
// If so, go through the loop again
|
|
stream.backUp(1);
|
|
return tokenString(stream, state);
|
|
}
|
|
// Are we in a definition (class, node, define)?
|
|
if (state.inDefinition) {
|
|
// If so, return def (i.e. for 'class myclass {' ; 'myclass' would be matched)
|
|
if (stream.match(/(\s+)?[\w:_]+(\s+)?/)) {
|
|
return 'def';
|
|
}
|
|
// Match the rest it the next time around
|
|
stream.match(/\s+{/);
|
|
state.inDefinition = false;
|
|
}
|
|
// Are we in an 'include' statement?
|
|
if (state.inInclude) {
|
|
// Match and return the included class
|
|
stream.match(/(\s+)?\S+(\s+)?/);
|
|
state.inInclude = false;
|
|
return 'def';
|
|
}
|
|
// Do we just have a function on our hands?
|
|
// In 'ensure_resource("myclass")', 'ensure_resource' is matched
|
|
if (stream.match(/(\s+)?\w+\(/)) {
|
|
stream.backUp(1);
|
|
return 'def';
|
|
}
|
|
// Have we matched the prior attribute regex?
|
|
if (attribute) {
|
|
stream.match(/(\s+)?\w+/);
|
|
return 'tag';
|
|
}
|
|
// Do we have Puppet specific words?
|
|
if (word && words.hasOwnProperty(word)) {
|
|
// Negates the initial next()
|
|
stream.backUp(1);
|
|
// rs move the stream
|
|
stream.match(/[\w]+/);
|
|
// We want to process these words differently
|
|
// do to the importance they have in Puppet
|
|
if (stream.match(/\s+\S+\s+{/, false)) {
|
|
state.inDefinition = true;
|
|
}
|
|
if (word == 'include') {
|
|
state.inInclude = true;
|
|
}
|
|
// Returns their value as state in the prior define methods
|
|
return words[word];
|
|
}
|
|
// Is there a match on a reference?
|
|
if (/(^|\s+)[A-Z][\w:_]+/.test(word)) {
|
|
// Negate the next()
|
|
stream.backUp(1);
|
|
// Match the full reference
|
|
stream.match(/(^|\s+)[A-Z][\w:_]+/);
|
|
return 'def';
|
|
}
|
|
// Have we matched the prior resource regex?
|
|
if (resource) {
|
|
stream.match(/(\s+)?[\w:_]+/);
|
|
return 'def';
|
|
}
|
|
// Have we matched the prior special_resource regex?
|
|
if (special_resource) {
|
|
stream.match(/(\s+)?[@]{1,2}/);
|
|
return 'special';
|
|
}
|
|
// Match all the comments. All of them.
|
|
if (ch == "#") {
|
|
stream.skipToEnd();
|
|
return "comment";
|
|
}
|
|
// Have we found a string?
|
|
if (ch == "'" || ch == '"') {
|
|
// Store the type (single or double)
|
|
state.pending = ch;
|
|
// Perform the looping function to find the end
|
|
return tokenString(stream, state);
|
|
}
|
|
// Match all the brackets
|
|
if (ch == '{' || ch == '}') {
|
|
return 'bracket';
|
|
}
|
|
// Match characters that we are going to assume
|
|
// are trying to be regex
|
|
if (ch == '/') {
|
|
stream.match(/.*?\//);
|
|
return 'variable-3';
|
|
}
|
|
// Match all the numbers
|
|
if (ch.match(/[0-9]/)) {
|
|
stream.eatWhile(/[0-9]+/);
|
|
return 'number';
|
|
}
|
|
// Match the '=' and '=>' operators
|
|
if (ch == '=') {
|
|
if (stream.peek() == '>') {
|
|
stream.next();
|
|
}
|
|
return "operator";
|
|
}
|
|
// Keep advancing through all the rest
|
|
stream.eatWhile(/[\w-]/);
|
|
// Return a blank line for everything else
|
|
return null;
|
|
}
|
|
// Start it all
|
|
return {
|
|
startState: function () {
|
|
var state = {};
|
|
state.inDefinition = false;
|
|
state.inInclude = false;
|
|
state.continueString = false;
|
|
state.pending = false;
|
|
return state;
|
|
},
|
|
token: function (stream, state) {
|
|
// Strip the spaces, but regex will account for them eitherway
|
|
if (stream.eatSpace()) return null;
|
|
// Go through the main process
|
|
return tokenize(stream, state);
|
|
}
|
|
};
|
|
});
|
|
|
|
CodeMirror.defineMIME("text/x-puppet", "puppet");
|
|
|
|
});
|