mirror of
				https://github.com/TriliumNext/Notes.git
				synced 2025-10-26 17:41:34 +08:00 
			
		
		
		
	
		
			
				
	
	
		
			434 lines
		
	
	
		
			12 KiB
		
	
	
	
		
			JavaScript
		
	
	
	
	
	
			
		
		
	
	
			434 lines
		
	
	
		
			12 KiB
		
	
	
	
		
			JavaScript
		
	
	
	
	
	
| // CodeMirror, copyright (c) by Marijn Haverbeke and others
 | |
| // Distributed under an MIT license: https://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("crystal", function(config) {
 | |
|     function wordRegExp(words, end) {
 | |
|       return new RegExp((end ? "" : "^") + "(?:" + words.join("|") + ")" + (end ? "$" : "\\b"));
 | |
|     }
 | |
| 
 | |
|     function chain(tokenize, stream, state) {
 | |
|       state.tokenize.push(tokenize);
 | |
|       return tokenize(stream, state);
 | |
|     }
 | |
| 
 | |
|     var operators = /^(?:[-+/%|&^]|\*\*?|[<>]{2})/;
 | |
|     var conditionalOperators = /^(?:[=!]~|===|<=>|[<>=!]=?|[|&]{2}|~)/;
 | |
|     var indexingOperators = /^(?:\[\][?=]?)/;
 | |
|     var anotherOperators = /^(?:\.(?:\.{2})?|->|[?:])/;
 | |
|     var idents = /^[a-z_\u009F-\uFFFF][a-zA-Z0-9_\u009F-\uFFFF]*/;
 | |
|     var types = /^[A-Z_\u009F-\uFFFF][a-zA-Z0-9_\u009F-\uFFFF]*/;
 | |
|     var keywords = wordRegExp([
 | |
|       "abstract", "alias", "as", "asm", "begin", "break", "case", "class", "def", "do",
 | |
|       "else", "elsif", "end", "ensure", "enum", "extend", "for", "fun", "if",
 | |
|       "include", "instance_sizeof", "lib", "macro", "module", "next", "of", "out", "pointerof",
 | |
|       "private", "protected", "rescue", "return", "require", "select", "sizeof", "struct",
 | |
|       "super", "then", "type", "typeof", "uninitialized", "union", "unless", "until", "when", "while", "with",
 | |
|       "yield", "__DIR__", "__END_LINE__", "__FILE__", "__LINE__"
 | |
|     ]);
 | |
|     var atomWords = wordRegExp(["true", "false", "nil", "self"]);
 | |
|     var indentKeywordsArray = [
 | |
|       "def", "fun", "macro",
 | |
|       "class", "module", "struct", "lib", "enum", "union",
 | |
|       "do", "for"
 | |
|     ];
 | |
|     var indentKeywords = wordRegExp(indentKeywordsArray);
 | |
|     var indentExpressionKeywordsArray = ["if", "unless", "case", "while", "until", "begin", "then"];
 | |
|     var indentExpressionKeywords = wordRegExp(indentExpressionKeywordsArray);
 | |
|     var dedentKeywordsArray = ["end", "else", "elsif", "rescue", "ensure"];
 | |
|     var dedentKeywords = wordRegExp(dedentKeywordsArray);
 | |
|     var dedentPunctualsArray = ["\\)", "\\}", "\\]"];
 | |
|     var dedentPunctuals = new RegExp("^(?:" + dedentPunctualsArray.join("|") + ")$");
 | |
|     var nextTokenizer = {
 | |
|       "def": tokenFollowIdent, "fun": tokenFollowIdent, "macro": tokenMacroDef,
 | |
|       "class": tokenFollowType, "module": tokenFollowType, "struct": tokenFollowType,
 | |
|       "lib": tokenFollowType, "enum": tokenFollowType, "union": tokenFollowType
 | |
|     };
 | |
|     var matching = {"[": "]", "{": "}", "(": ")", "<": ">"};
 | |
| 
 | |
|     function tokenBase(stream, state) {
 | |
|       if (stream.eatSpace()) {
 | |
|         return null;
 | |
|       }
 | |
| 
 | |
|       // Macros
 | |
|       if (state.lastToken != "\\" && stream.match("{%", false)) {
 | |
|         return chain(tokenMacro("%", "%"), stream, state);
 | |
|       }
 | |
| 
 | |
|       if (state.lastToken != "\\" && stream.match("{{", false)) {
 | |
|         return chain(tokenMacro("{", "}"), stream, state);
 | |
|       }
 | |
| 
 | |
|       // Comments
 | |
|       if (stream.peek() == "#") {
 | |
|         stream.skipToEnd();
 | |
|         return "comment";
 | |
|       }
 | |
| 
 | |
|       // Variables and keywords
 | |
|       var matched;
 | |
|       if (stream.match(idents)) {
 | |
|         stream.eat(/[?!]/);
 | |
| 
 | |
|         matched = stream.current();
 | |
|         if (stream.eat(":")) {
 | |
|           return "atom";
 | |
|         } else if (state.lastToken == ".") {
 | |
|           return "property";
 | |
|         } else if (keywords.test(matched)) {
 | |
|           if (indentKeywords.test(matched)) {
 | |
|             if (!(matched == "fun" && state.blocks.indexOf("lib") >= 0) && !(matched == "def" && state.lastToken == "abstract")) {
 | |
|               state.blocks.push(matched);
 | |
|               state.currentIndent += 1;
 | |
|             }
 | |
|           } else if ((state.lastStyle == "operator" || !state.lastStyle) && indentExpressionKeywords.test(matched)) {
 | |
|             state.blocks.push(matched);
 | |
|             state.currentIndent += 1;
 | |
|           } else if (matched == "end") {
 | |
|             state.blocks.pop();
 | |
|             state.currentIndent -= 1;
 | |
|           }
 | |
| 
 | |
|           if (nextTokenizer.hasOwnProperty(matched)) {
 | |
|             state.tokenize.push(nextTokenizer[matched]);
 | |
|           }
 | |
| 
 | |
|           return "keyword";
 | |
|         } else if (atomWords.test(matched)) {
 | |
|           return "atom";
 | |
|         }
 | |
| 
 | |
|         return "variable";
 | |
|       }
 | |
| 
 | |
|       // Class variables and instance variables
 | |
|       // or attributes
 | |
|       if (stream.eat("@")) {
 | |
|         if (stream.peek() == "[") {
 | |
|           return chain(tokenNest("[", "]", "meta"), stream, state);
 | |
|         }
 | |
| 
 | |
|         stream.eat("@");
 | |
|         stream.match(idents) || stream.match(types);
 | |
|         return "variable-2";
 | |
|       }
 | |
| 
 | |
|       // Constants and types
 | |
|       if (stream.match(types)) {
 | |
|         return "tag";
 | |
|       }
 | |
| 
 | |
|       // Symbols or ':' operator
 | |
|       if (stream.eat(":")) {
 | |
|         if (stream.eat("\"")) {
 | |
|           return chain(tokenQuote("\"", "atom", false), stream, state);
 | |
|         } else if (stream.match(idents) || stream.match(types) ||
 | |
|                    stream.match(operators) || stream.match(conditionalOperators) || stream.match(indexingOperators)) {
 | |
|           return "atom";
 | |
|         }
 | |
|         stream.eat(":");
 | |
|         return "operator";
 | |
|       }
 | |
| 
 | |
|       // Strings
 | |
|       if (stream.eat("\"")) {
 | |
|         return chain(tokenQuote("\"", "string", true), stream, state);
 | |
|       }
 | |
| 
 | |
|       // Strings or regexps or macro variables or '%' operator
 | |
|       if (stream.peek() == "%") {
 | |
|         var style = "string";
 | |
|         var embed = true;
 | |
|         var delim;
 | |
| 
 | |
|         if (stream.match("%r")) {
 | |
|           // Regexps
 | |
|           style = "string-2";
 | |
|           delim = stream.next();
 | |
|         } else if (stream.match("%w")) {
 | |
|           embed = false;
 | |
|           delim = stream.next();
 | |
|         } else if (stream.match("%q")) {
 | |
|           embed = false;
 | |
|           delim = stream.next();
 | |
|         } else {
 | |
|           if(delim = stream.match(/^%([^\w\s=])/)) {
 | |
|             delim = delim[1];
 | |
|           } else if (stream.match(/^%[a-zA-Z0-9_\u009F-\uFFFF]*/)) {
 | |
|             // Macro variables
 | |
|             return "meta";
 | |
|           } else {
 | |
|             // '%' operator
 | |
|             return "operator";
 | |
|           }
 | |
|         }
 | |
| 
 | |
|         if (matching.hasOwnProperty(delim)) {
 | |
|           delim = matching[delim];
 | |
|         }
 | |
|         return chain(tokenQuote(delim, style, embed), stream, state);
 | |
|       }
 | |
| 
 | |
|       // Here Docs
 | |
|       if (matched = stream.match(/^<<-('?)([A-Z]\w*)\1/)) {
 | |
|         return chain(tokenHereDoc(matched[2], !matched[1]), stream, state)
 | |
|       }
 | |
| 
 | |
|       // Characters
 | |
|       if (stream.eat("'")) {
 | |
|         stream.match(/^(?:[^']|\\(?:[befnrtv0'"]|[0-7]{3}|u(?:[0-9a-fA-F]{4}|\{[0-9a-fA-F]{1,6}\})))/);
 | |
|         stream.eat("'");
 | |
|         return "atom";
 | |
|       }
 | |
| 
 | |
|       // Numbers
 | |
|       if (stream.eat("0")) {
 | |
|         if (stream.eat("x")) {
 | |
|           stream.match(/^[0-9a-fA-F]+/);
 | |
|         } else if (stream.eat("o")) {
 | |
|           stream.match(/^[0-7]+/);
 | |
|         } else if (stream.eat("b")) {
 | |
|           stream.match(/^[01]+/);
 | |
|         }
 | |
|         return "number";
 | |
|       }
 | |
| 
 | |
|       if (stream.eat(/^\d/)) {
 | |
|         stream.match(/^\d*(?:\.\d+)?(?:[eE][+-]?\d+)?/);
 | |
|         return "number";
 | |
|       }
 | |
| 
 | |
|       // Operators
 | |
|       if (stream.match(operators)) {
 | |
|         stream.eat("="); // Operators can follow assign symbol.
 | |
|         return "operator";
 | |
|       }
 | |
| 
 | |
|       if (stream.match(conditionalOperators) || stream.match(anotherOperators)) {
 | |
|         return "operator";
 | |
|       }
 | |
| 
 | |
|       // Parens and braces
 | |
|       if (matched = stream.match(/[({[]/, false)) {
 | |
|         matched = matched[0];
 | |
|         return chain(tokenNest(matched, matching[matched], null), stream, state);
 | |
|       }
 | |
| 
 | |
|       // Escapes
 | |
|       if (stream.eat("\\")) {
 | |
|         stream.next();
 | |
|         return "meta";
 | |
|       }
 | |
| 
 | |
|       stream.next();
 | |
|       return null;
 | |
|     }
 | |
| 
 | |
|     function tokenNest(begin, end, style, started) {
 | |
|       return function (stream, state) {
 | |
|         if (!started && stream.match(begin)) {
 | |
|           state.tokenize[state.tokenize.length - 1] = tokenNest(begin, end, style, true);
 | |
|           state.currentIndent += 1;
 | |
|           return style;
 | |
|         }
 | |
| 
 | |
|         var nextStyle = tokenBase(stream, state);
 | |
|         if (stream.current() === end) {
 | |
|           state.tokenize.pop();
 | |
|           state.currentIndent -= 1;
 | |
|           nextStyle = style;
 | |
|         }
 | |
| 
 | |
|         return nextStyle;
 | |
|       };
 | |
|     }
 | |
| 
 | |
|     function tokenMacro(begin, end, started) {
 | |
|       return function (stream, state) {
 | |
|         if (!started && stream.match("{" + begin)) {
 | |
|           state.currentIndent += 1;
 | |
|           state.tokenize[state.tokenize.length - 1] = tokenMacro(begin, end, true);
 | |
|           return "meta";
 | |
|         }
 | |
| 
 | |
|         if (stream.match(end + "}")) {
 | |
|           state.currentIndent -= 1;
 | |
|           state.tokenize.pop();
 | |
|           return "meta";
 | |
|         }
 | |
| 
 | |
|         return tokenBase(stream, state);
 | |
|       };
 | |
|     }
 | |
| 
 | |
|     function tokenMacroDef(stream, state) {
 | |
|       if (stream.eatSpace()) {
 | |
|         return null;
 | |
|       }
 | |
| 
 | |
|       var matched;
 | |
|       if (matched = stream.match(idents)) {
 | |
|         if (matched == "def") {
 | |
|           return "keyword";
 | |
|         }
 | |
|         stream.eat(/[?!]/);
 | |
|       }
 | |
| 
 | |
|       state.tokenize.pop();
 | |
|       return "def";
 | |
|     }
 | |
| 
 | |
|     function tokenFollowIdent(stream, state) {
 | |
|       if (stream.eatSpace()) {
 | |
|         return null;
 | |
|       }
 | |
| 
 | |
|       if (stream.match(idents)) {
 | |
|         stream.eat(/[!?]/);
 | |
|       } else {
 | |
|         stream.match(operators) || stream.match(conditionalOperators) || stream.match(indexingOperators);
 | |
|       }
 | |
|       state.tokenize.pop();
 | |
|       return "def";
 | |
|     }
 | |
| 
 | |
|     function tokenFollowType(stream, state) {
 | |
|       if (stream.eatSpace()) {
 | |
|         return null;
 | |
|       }
 | |
| 
 | |
|       stream.match(types);
 | |
|       state.tokenize.pop();
 | |
|       return "def";
 | |
|     }
 | |
| 
 | |
|     function tokenQuote(end, style, embed) {
 | |
|       return function (stream, state) {
 | |
|         var escaped = false;
 | |
| 
 | |
|         while (stream.peek()) {
 | |
|           if (!escaped) {
 | |
|             if (stream.match("{%", false)) {
 | |
|               state.tokenize.push(tokenMacro("%", "%"));
 | |
|               return style;
 | |
|             }
 | |
| 
 | |
|             if (stream.match("{{", false)) {
 | |
|               state.tokenize.push(tokenMacro("{", "}"));
 | |
|               return style;
 | |
|             }
 | |
| 
 | |
|             if (embed && stream.match("#{", false)) {
 | |
|               state.tokenize.push(tokenNest("#{", "}", "meta"));
 | |
|               return style;
 | |
|             }
 | |
| 
 | |
|             var ch = stream.next();
 | |
| 
 | |
|             if (ch == end) {
 | |
|               state.tokenize.pop();
 | |
|               return style;
 | |
|             }
 | |
| 
 | |
|             escaped = embed && ch == "\\";
 | |
|           } else {
 | |
|             stream.next();
 | |
|             escaped = false;
 | |
|           }
 | |
|         }
 | |
| 
 | |
|         return style;
 | |
|       };
 | |
|     }
 | |
| 
 | |
|     function tokenHereDoc(phrase, embed) {
 | |
|       return function (stream, state) {
 | |
|         if (stream.sol()) {
 | |
|           stream.eatSpace()
 | |
|           if (stream.match(phrase)) {
 | |
|             state.tokenize.pop();
 | |
|             return "string";
 | |
|           }
 | |
|         }
 | |
| 
 | |
|         var escaped = false;
 | |
|         while (stream.peek()) {
 | |
|           if (!escaped) {
 | |
|             if (stream.match("{%", false)) {
 | |
|               state.tokenize.push(tokenMacro("%", "%"));
 | |
|               return "string";
 | |
|             }
 | |
| 
 | |
|             if (stream.match("{{", false)) {
 | |
|               state.tokenize.push(tokenMacro("{", "}"));
 | |
|               return "string";
 | |
|             }
 | |
| 
 | |
|             if (embed && stream.match("#{", false)) {
 | |
|               state.tokenize.push(tokenNest("#{", "}", "meta"));
 | |
|               return "string";
 | |
|             }
 | |
| 
 | |
|             escaped = embed && stream.next() == "\\";
 | |
|           } else {
 | |
|             stream.next();
 | |
|             escaped = false;
 | |
|           }
 | |
|         }
 | |
| 
 | |
|         return "string";
 | |
|       }
 | |
|     }
 | |
| 
 | |
|     return {
 | |
|       startState: function () {
 | |
|         return {
 | |
|           tokenize: [tokenBase],
 | |
|           currentIndent: 0,
 | |
|           lastToken: null,
 | |
|           lastStyle: null,
 | |
|           blocks: []
 | |
|         };
 | |
|       },
 | |
| 
 | |
|       token: function (stream, state) {
 | |
|         var style = state.tokenize[state.tokenize.length - 1](stream, state);
 | |
|         var token = stream.current();
 | |
| 
 | |
|         if (style && style != "comment") {
 | |
|           state.lastToken = token;
 | |
|           state.lastStyle = style;
 | |
|         }
 | |
| 
 | |
|         return style;
 | |
|       },
 | |
| 
 | |
|       indent: function (state, textAfter) {
 | |
|         textAfter = textAfter.replace(/^\s*(?:\{%)?\s*|\s*(?:%\})?\s*$/g, "");
 | |
| 
 | |
|         if (dedentKeywords.test(textAfter) || dedentPunctuals.test(textAfter)) {
 | |
|           return config.indentUnit * (state.currentIndent - 1);
 | |
|         }
 | |
| 
 | |
|         return config.indentUnit * state.currentIndent;
 | |
|       },
 | |
| 
 | |
|       fold: "indent",
 | |
|       electricInput: wordRegExp(dedentPunctualsArray.concat(dedentKeywordsArray), true),
 | |
|       lineComment: '#'
 | |
|     };
 | |
|   });
 | |
| 
 | |
|   CodeMirror.defineMIME("text/x-crystal", "crystal");
 | |
| });
 | 
