##// END OF EJS Templates
components: small order of elements change and a cleanup
components: small order of elements change and a cleanup

File last commit:

r346:1a48416d default
r983:c5d592d8 default
Show More
cypher.js
146 lines | 6.1 KiB | application/javascript | JavascriptLexer
project: added all source files and assets
r1 // CodeMirror, copyright (c) by Marijn Haverbeke and others
// Distributed under an MIT license: http://codemirror.net/LICENSE
// By the Neo4j Team and contributors.
// https://github.com/neo4j-contrib/CodeMirror
(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";
var wordRegexp = function(words) {
return new RegExp("^(?:" + words.join("|") + ")$", "i");
};
CodeMirror.defineMode("cypher", function(config) {
var tokenBase = function(stream/*, state*/) {
var ch = stream.next();
if (ch === "\"" || ch === "'") {
stream.match(/.+?["']/);
return "string";
}
if (/[{}\(\),\.;\[\]]/.test(ch)) {
curPunc = ch;
return "node";
} else if (ch === "/" && stream.eat("/")) {
stream.skipToEnd();
return "comment";
} else if (operatorChars.test(ch)) {
stream.eatWhile(operatorChars);
return null;
} else {
stream.eatWhile(/[_\w\d]/);
if (stream.eat(":")) {
stream.eatWhile(/[\w\d_\-]/);
return "atom";
}
var word = stream.current();
if (funcs.test(word)) return "builtin";
if (preds.test(word)) return "def";
if (keywords.test(word)) return "keyword";
return "variable";
}
};
var pushContext = function(state, type, col) {
return state.context = {
prev: state.context,
indent: state.indent,
col: col,
type: type
};
};
var popContext = function(state) {
state.indent = state.context.indent;
return state.context = state.context.prev;
};
var indentUnit = config.indentUnit;
var curPunc;
codemirror: update from 5.4.0 to 5.11.0, fixes #3154
r346 var funcs = wordRegexp(["abs", "acos", "allShortestPaths", "asin", "atan", "atan2", "avg", "ceil", "coalesce", "collect", "cos", "cot", "count", "degrees", "e", "endnode", "exp", "extract", "filter", "floor", "haversin", "head", "id", "keys", "labels", "last", "left", "length", "log", "log10", "lower", "ltrim", "max", "min", "node", "nodes", "percentileCont", "percentileDisc", "pi", "radians", "rand", "range", "reduce", "rel", "relationship", "relationships", "replace", "reverse", "right", "round", "rtrim", "shortestPath", "sign", "sin", "size", "split", "sqrt", "startnode", "stdev", "stdevp", "str", "substring", "sum", "tail", "tan", "timestamp", "toFloat", "toInt", "toString", "trim", "type", "upper"]);
var preds = wordRegexp(["all", "and", "any", "contains", "exists", "has", "in", "none", "not", "or", "single", "xor"]);
var keywords = wordRegexp(["as", "asc", "ascending", "assert", "by", "case", "commit", "constraint", "create", "csv", "cypher", "delete", "desc", "descending", "detach", "distinct", "drop", "else", "end", "ends", "explain", "false", "fieldterminator", "foreach", "from", "headers", "in", "index", "is", "join", "limit", "load", "match", "merge", "null", "on", "optional", "order", "periodic", "profile", "remove", "return", "scan", "set", "skip", "start", "starts", "then", "true", "union", "unique", "unwind", "using", "when", "where", "with"]);
project: added all source files and assets
r1 var operatorChars = /[*+\-<>=&|~%^]/;
return {
startState: function(/*base*/) {
return {
tokenize: tokenBase,
context: null,
indent: 0,
col: 0
};
},
token: function(stream, state) {
if (stream.sol()) {
if (state.context && (state.context.align == null)) {
state.context.align = false;
}
state.indent = stream.indentation();
}
if (stream.eatSpace()) {
return null;
}
var style = state.tokenize(stream, state);
if (style !== "comment" && state.context && (state.context.align == null) && state.context.type !== "pattern") {
state.context.align = true;
}
if (curPunc === "(") {
pushContext(state, ")", stream.column());
} else if (curPunc === "[") {
pushContext(state, "]", stream.column());
} else if (curPunc === "{") {
pushContext(state, "}", stream.column());
} else if (/[\]\}\)]/.test(curPunc)) {
while (state.context && state.context.type === "pattern") {
popContext(state);
}
if (state.context && curPunc === state.context.type) {
popContext(state);
}
} else if (curPunc === "." && state.context && state.context.type === "pattern") {
popContext(state);
} else if (/atom|string|variable/.test(style) && state.context) {
if (/[\}\]]/.test(state.context.type)) {
pushContext(state, "pattern", stream.column());
} else if (state.context.type === "pattern" && !state.context.align) {
state.context.align = true;
state.context.col = stream.column();
}
}
return style;
},
indent: function(state, textAfter) {
var firstChar = textAfter && textAfter.charAt(0);
var context = state.context;
if (/[\]\}]/.test(firstChar)) {
while (context && context.type === "pattern") {
context = context.prev;
}
}
var closing = context && firstChar === context.type;
if (!context) return 0;
if (context.type === "keywords") return CodeMirror.commands.newlineAndIndent;
if (context.align) return context.col + (closing ? 0 : 1);
return context.indent + (closing ? 0 : indentUnit);
}
};
});
CodeMirror.modeExtensions["cypher"] = {
autoFormatLineBreaks: function(text) {
var i, lines, reProcessedPortion;
var lines = text.split("\n");
var reProcessedPortion = /\s+\b(return|where|order by|match|with|skip|limit|create|delete|set)\b\s/g;
for (var i = 0; i < lines.length; i++)
lines[i] = lines[i].replace(reProcessedPortion, " \n$1 ").trim();
return lines.join("\n");
}
};
CodeMirror.defineMIME("application/x-cypher-query", "cypher");
});