overleaf/services/web/public/js/ace-1.4.5/mode-elixir.js
Alasdair Smith 350e2ac38f Merge pull request #1886 from overleaf/as-upgrade-ace-1.4.5
Upgrade to Ace v1.4.5

GitOrigin-RevId: 7ab7ba24c0bfc927567fbd4196e05a1c3f26d283
2019-06-20 09:21:27 +00:00

502 lines
No EOL
23 KiB
JavaScript

ace.define("ace/mode/elixir_highlight_rules",[], function(require, exports, module) {
"use strict";
var oop = require("../lib/oop");
var TextHighlightRules = require("./text_highlight_rules").TextHighlightRules;
var ElixirHighlightRules = function() {
this.$rules = { start:
[ { token:
[ 'meta.module.elixir',
'keyword.control.module.elixir',
'meta.module.elixir',
'entity.name.type.module.elixir' ],
regex: '^(\\s*)(defmodule)(\\s+)((?:[A-Z]\\w*\\s*\\.\\s*)*[A-Z]\\w*)' },
{ token: 'comment.documentation.heredoc',
regex: '@(?:module|type)?doc (?:~[a-z])?"""',
push:
[ { token: 'comment.documentation.heredoc',
regex: '\\s*"""',
next: 'pop' },
{ include: '#interpolated_elixir' },
{ include: '#escaped_char' },
{ defaultToken: 'comment.documentation.heredoc' } ],
comment: '@doc with heredocs is treated as documentation' },
{ token: 'comment.documentation.heredoc',
regex: '@(?:module|type)?doc ~[A-Z]"""',
push:
[ { token: 'comment.documentation.heredoc',
regex: '\\s*"""',
next: 'pop' },
{ defaultToken: 'comment.documentation.heredoc' } ],
comment: '@doc with heredocs is treated as documentation' },
{ token: 'comment.documentation.heredoc',
regex: '@(?:module|type)?doc (?:~[a-z])?\'\'\'',
push:
[ { token: 'comment.documentation.heredoc',
regex: '\\s*\'\'\'',
next: 'pop' },
{ include: '#interpolated_elixir' },
{ include: '#escaped_char' },
{ defaultToken: 'comment.documentation.heredoc' } ],
comment: '@doc with heredocs is treated as documentation' },
{ token: 'comment.documentation.heredoc',
regex: '@(?:module|type)?doc ~[A-Z]\'\'\'',
push:
[ { token: 'comment.documentation.heredoc',
regex: '\\s*\'\'\'',
next: 'pop' },
{ defaultToken: 'comment.documentation.heredoc' } ],
comment: '@doc with heredocs is treated as documentation' },
{ token: 'comment.documentation.false',
regex: '@(?:module|type)?doc false',
comment: '@doc false is treated as documentation' },
{ token: 'comment.documentation.string',
regex: '@(?:module|type)?doc "',
push:
[ { token: 'comment.documentation.string',
regex: '"',
next: 'pop' },
{ include: '#interpolated_elixir' },
{ include: '#escaped_char' },
{ defaultToken: 'comment.documentation.string' } ],
comment: '@doc with string is treated as documentation' },
{ token: 'keyword.control.elixir',
regex: '\\b(?:do|end|case|bc|lc|for|if|cond|unless|try|receive|fn|defmodule|defp?|defprotocol|defimpl|defrecord|defstruct|defmacrop?|defdelegate|defcallback|defmacrocallback|defexception|defoverridable|exit|after|rescue|catch|else|raise|throw|import|require|alias|use|quote|unquote|super)\\b(?![?!])',
TODO: 'FIXME: regexp doesn\'t have js equivalent',
originalRegex: '(?<!\\.)\\b(do|end|case|bc|lc|for|if|cond|unless|try|receive|fn|defmodule|defp?|defprotocol|defimpl|defrecord|defstruct|defmacrop?|defdelegate|defcallback|defmacrocallback|defexception|defoverridable|exit|after|rescue|catch|else|raise|throw|import|require|alias|use|quote|unquote|super)\\b(?![?!])' },
{ token: 'keyword.operator.elixir',
regex: '\\b(?:and|not|or|when|xor|in|inlist|inbits)\\b',
TODO: 'FIXME: regexp doesn\'t have js equivalent',
originalRegex: '(?<!\\.)\\b(and|not|or|when|xor|in|inlist|inbits)\\b',
comment: ' as above, just doesn\'t need a \'end\' and does a logic operation' },
{ token: 'constant.language.elixir',
regex: '\\b(?:nil|true|false)\\b(?![?!])' },
{ token: 'variable.language.elixir',
regex: '\\b__(?:CALLER|ENV|MODULE|DIR)__\\b(?![?!])' },
{ token:
[ 'punctuation.definition.variable.elixir',
'variable.other.readwrite.module.elixir' ],
regex: '(@)([a-zA-Z_]\\w*)' },
{ token:
[ 'punctuation.definition.variable.elixir',
'variable.other.anonymous.elixir' ],
regex: '(&)(\\d*)' },
{ token: 'variable.other.constant.elixir',
regex: '\\b[A-Z]\\w*\\b' },
{ token: 'constant.numeric.elixir',
regex: '\\b(?:0x[\\da-fA-F](?:_?[\\da-fA-F])*|\\d(?:_?\\d)*(?:\\.(?![^[:space:][:digit:]])(?:_?\\d)*)?(?:[eE][-+]?\\d(?:_?\\d)*)?|0b[01]+|0o[0-7]+)\\b',
TODO: 'FIXME: regexp doesn\'t have js equivalent',
originalRegex: '\\b(0x\\h(?>_?\\h)*|\\d(?>_?\\d)*(\\.(?![^[:space:][:digit:]])(?>_?\\d)*)?([eE][-+]?\\d(?>_?\\d)*)?|0b[01]+|0o[0-7]+)\\b' },
{ token: 'punctuation.definition.constant.elixir',
regex: ':\'',
push:
[ { token: 'punctuation.definition.constant.elixir',
regex: '\'',
next: 'pop' },
{ include: '#interpolated_elixir' },
{ include: '#escaped_char' },
{ defaultToken: 'constant.other.symbol.single-quoted.elixir' } ] },
{ token: 'punctuation.definition.constant.elixir',
regex: ':"',
push:
[ { token: 'punctuation.definition.constant.elixir',
regex: '"',
next: 'pop' },
{ include: '#interpolated_elixir' },
{ include: '#escaped_char' },
{ defaultToken: 'constant.other.symbol.double-quoted.elixir' } ] },
{ token: 'punctuation.definition.string.begin.elixir',
regex: '(?:\'\'\')',
TODO: 'FIXME: regexp doesn\'t have js equivalent',
originalRegex: '(?>\'\'\')',
push:
[ { token: 'punctuation.definition.string.end.elixir',
regex: '^\\s*\'\'\'',
next: 'pop' },
{ include: '#interpolated_elixir' },
{ include: '#escaped_char' },
{ defaultToken: 'support.function.variable.quoted.single.heredoc.elixir' } ],
comment: 'Single-quoted heredocs' },
{ token: 'punctuation.definition.string.begin.elixir',
regex: '\'',
push:
[ { token: 'punctuation.definition.string.end.elixir',
regex: '\'',
next: 'pop' },
{ include: '#interpolated_elixir' },
{ include: '#escaped_char' },
{ defaultToken: 'support.function.variable.quoted.single.elixir' } ],
comment: 'single quoted string (allows for interpolation)' },
{ token: 'punctuation.definition.string.begin.elixir',
regex: '(?:""")',
TODO: 'FIXME: regexp doesn\'t have js equivalent',
originalRegex: '(?>""")',
push:
[ { token: 'punctuation.definition.string.end.elixir',
regex: '^\\s*"""',
next: 'pop' },
{ include: '#interpolated_elixir' },
{ include: '#escaped_char' },
{ defaultToken: 'string.quoted.double.heredoc.elixir' } ],
comment: 'Double-quoted heredocs' },
{ token: 'punctuation.definition.string.begin.elixir',
regex: '"',
push:
[ { token: 'punctuation.definition.string.end.elixir',
regex: '"',
next: 'pop' },
{ include: '#interpolated_elixir' },
{ include: '#escaped_char' },
{ defaultToken: 'string.quoted.double.elixir' } ],
comment: 'double quoted string (allows for interpolation)' },
{ token: 'punctuation.definition.string.begin.elixir',
regex: '~[a-z](?:""")',
TODO: 'FIXME: regexp doesn\'t have js equivalent',
originalRegex: '~[a-z](?>""")',
push:
[ { token: 'punctuation.definition.string.end.elixir',
regex: '^\\s*"""',
next: 'pop' },
{ include: '#interpolated_elixir' },
{ include: '#escaped_char' },
{ defaultToken: 'string.quoted.double.heredoc.elixir' } ],
comment: 'Double-quoted heredocs sigils' },
{ token: 'punctuation.definition.string.begin.elixir',
regex: '~[a-z]\\{',
push:
[ { token: 'punctuation.definition.string.end.elixir',
regex: '\\}[a-z]*',
next: 'pop' },
{ include: '#interpolated_elixir' },
{ include: '#escaped_char' },
{ defaultToken: 'string.interpolated.elixir' } ],
comment: 'sigil (allow for interpolation)' },
{ token: 'punctuation.definition.string.begin.elixir',
regex: '~[a-z]\\[',
push:
[ { token: 'punctuation.definition.string.end.elixir',
regex: '\\][a-z]*',
next: 'pop' },
{ include: '#interpolated_elixir' },
{ include: '#escaped_char' },
{ defaultToken: 'string.interpolated.elixir' } ],
comment: 'sigil (allow for interpolation)' },
{ token: 'punctuation.definition.string.begin.elixir',
regex: '~[a-z]\\<',
push:
[ { token: 'punctuation.definition.string.end.elixir',
regex: '\\>[a-z]*',
next: 'pop' },
{ include: '#interpolated_elixir' },
{ include: '#escaped_char' },
{ defaultToken: 'string.interpolated.elixir' } ],
comment: 'sigil (allow for interpolation)' },
{ token: 'punctuation.definition.string.begin.elixir',
regex: '~[a-z]\\(',
push:
[ { token: 'punctuation.definition.string.end.elixir',
regex: '\\)[a-z]*',
next: 'pop' },
{ include: '#interpolated_elixir' },
{ include: '#escaped_char' },
{ defaultToken: 'string.interpolated.elixir' } ],
comment: 'sigil (allow for interpolation)' },
{ token: 'punctuation.definition.string.begin.elixir',
regex: '~[a-z][^\\w]',
push:
[ { token: 'punctuation.definition.string.end.elixir',
regex: '[^\\w][a-z]*',
next: 'pop' },
{ include: '#interpolated_elixir' },
{ include: '#escaped_char' },
{ include: '#escaped_char' },
{ defaultToken: 'string.interpolated.elixir' } ],
comment: 'sigil (allow for interpolation)' },
{ token: 'punctuation.definition.string.begin.elixir',
regex: '~[A-Z](?:""")',
TODO: 'FIXME: regexp doesn\'t have js equivalent',
originalRegex: '~[A-Z](?>""")',
push:
[ { token: 'punctuation.definition.string.end.elixir',
regex: '^\\s*"""',
next: 'pop' },
{ defaultToken: 'string.quoted.other.literal.upper.elixir' } ],
comment: 'Double-quoted heredocs sigils' },
{ token: 'punctuation.definition.string.begin.elixir',
regex: '~[A-Z]\\{',
push:
[ { token: 'punctuation.definition.string.end.elixir',
regex: '\\}[a-z]*',
next: 'pop' },
{ defaultToken: 'string.quoted.other.literal.upper.elixir' } ],
comment: 'sigil (without interpolation)' },
{ token: 'punctuation.definition.string.begin.elixir',
regex: '~[A-Z]\\[',
push:
[ { token: 'punctuation.definition.string.end.elixir',
regex: '\\][a-z]*',
next: 'pop' },
{ defaultToken: 'string.quoted.other.literal.upper.elixir' } ],
comment: 'sigil (without interpolation)' },
{ token: 'punctuation.definition.string.begin.elixir',
regex: '~[A-Z]\\<',
push:
[ { token: 'punctuation.definition.string.end.elixir',
regex: '\\>[a-z]*',
next: 'pop' },
{ defaultToken: 'string.quoted.other.literal.upper.elixir' } ],
comment: 'sigil (without interpolation)' },
{ token: 'punctuation.definition.string.begin.elixir',
regex: '~[A-Z]\\(',
push:
[ { token: 'punctuation.definition.string.end.elixir',
regex: '\\)[a-z]*',
next: 'pop' },
{ defaultToken: 'string.quoted.other.literal.upper.elixir' } ],
comment: 'sigil (without interpolation)' },
{ token: 'punctuation.definition.string.begin.elixir',
regex: '~[A-Z][^\\w]',
push:
[ { token: 'punctuation.definition.string.end.elixir',
regex: '[^\\w][a-z]*',
next: 'pop' },
{ defaultToken: 'string.quoted.other.literal.upper.elixir' } ],
comment: 'sigil (without interpolation)' },
{ token: ['punctuation.definition.constant.elixir', 'constant.other.symbol.elixir'],
regex: '(:)([a-zA-Z_][\\w@]*(?:[?!]|=(?![>=]))?|\\<\\>|===?|!==?|<<>>|<<<|>>>|~~~|::|<\\-|\\|>|=>|~|~=|=|/|\\\\\\\\|\\*\\*?|\\.\\.?\\.?|>=?|<=?|&&?&?|\\+\\+?|\\-\\-?|\\|\\|?\\|?|\\!|@|\\%?\\{\\}|%|\\[\\]|\\^(?:\\^\\^)?)',
TODO: 'FIXME: regexp doesn\'t have js equivalent',
originalRegex: '(?<!:)(:)(?>[a-zA-Z_][\\w@]*(?>[?!]|=(?![>=]))?|\\<\\>|===?|!==?|<<>>|<<<|>>>|~~~|::|<\\-|\\|>|=>|~|~=|=|/|\\\\\\\\|\\*\\*?|\\.\\.?\\.?|>=?|<=?|&&?&?|\\+\\+?|\\-\\-?|\\|\\|?\\|?|\\!|@|\\%?\\{\\}|%|\\[\\]|\\^(\\^\\^)?)',
comment: 'symbols' },
{ token: 'punctuation.definition.constant.elixir',
regex: '(?:[a-zA-Z_][\\w@]*(?:[?!])?):(?!:)',
TODO: 'FIXME: regexp doesn\'t have js equivalent',
originalRegex: '(?>[a-zA-Z_][\\w@]*(?>[?!])?)(:)(?!:)',
comment: 'symbols' },
{ token:
[ 'punctuation.definition.comment.elixir',
'comment.line.number-sign.elixir' ],
regex: '(#)(.*)' },
{ token: 'constant.numeric.elixir',
regex: '\\?(?:\\\\(?:x[\\da-fA-F]{1,2}(?![\\da-fA-F])\\b|[^xMC])|[^\\s\\\\])',
TODO: 'FIXME: regexp doesn\'t have js equivalent',
originalRegex: '(?<!\\w)\\?(\\\\(x\\h{1,2}(?!\\h)\\b|[^xMC])|[^\\s\\\\])',
comment: '\n\t\t\tmatches questionmark-letters.\n\n\t\t\texamples (1st alternation = hex):\n\t\t\t?\\x1 ?\\x61\n\n\t\t\texamples (2rd alternation = escaped):\n\t\t\t?\\n ?\\b\n\n\t\t\texamples (3rd alternation = normal):\n\t\t\t?a ?A ?0 \n\t\t\t?* ?" ?( \n\t\t\t?. ?#\n\t\t\t\n\t\t\tthe negative lookbehind prevents against matching\n\t\t\tp(42.tainted?)\n\t\t\t' },
{ token: 'keyword.operator.assignment.augmented.elixir',
regex: '\\+=|\\-=|\\|\\|=|~=|&&=' },
{ token: 'keyword.operator.comparison.elixir',
regex: '===?|!==?|<=?|>=?' },
{ token: 'keyword.operator.bitwise.elixir',
regex: '\\|{3}|&{3}|\\^{3}|<{3}|>{3}|~{3}' },
{ token: 'keyword.operator.logical.elixir',
regex: '!+|\\bnot\\b|&&|\\band\\b|\\|\\||\\bor\\b|\\bxor\\b',
originalRegex: '(?<=[ \\t])!+|\\bnot\\b|&&|\\band\\b|\\|\\||\\bor\\b|\\bxor\\b' },
{ token: 'keyword.operator.arithmetic.elixir',
regex: '\\*|\\+|\\-|/' },
{ token: 'keyword.operator.other.elixir',
regex: '\\||\\+\\+|\\-\\-|\\*\\*|\\\\\\\\|\\<\\-|\\<\\>|\\<\\<|\\>\\>|\\:\\:|\\.\\.|\\|>|~|=>' },
{ token: 'keyword.operator.assignment.elixir', regex: '=' },
{ token: 'punctuation.separator.other.elixir', regex: ':' },
{ token: 'punctuation.separator.statement.elixir',
regex: '\\;' },
{ token: 'punctuation.separator.object.elixir', regex: ',' },
{ token: 'punctuation.separator.method.elixir', regex: '\\.' },
{ token: 'punctuation.section.scope.elixir', regex: '\\{|\\}' },
{ token: 'punctuation.section.array.elixir', regex: '\\[|\\]' },
{ token: 'punctuation.section.function.elixir',
regex: '\\(|\\)' } ],
'#escaped_char':
[ { token: 'constant.character.escape.elixir',
regex: '\\\\(?:x[\\da-fA-F]{1,2}|.)' } ],
'#interpolated_elixir':
[ { token:
[ 'source.elixir.embedded.source',
'source.elixir.embedded.source.empty' ],
regex: '(#\\{)(\\})' },
{ todo:
{ token: 'punctuation.section.embedded.elixir',
regex: '#\\{',
push:
[ { token: 'punctuation.section.embedded.elixir',
regex: '\\}',
next: 'pop' },
{ include: '#nest_curly_and_self' },
{ include: '$self' },
{ defaultToken: 'source.elixir.embedded.source' } ] } } ],
'#nest_curly_and_self':
[ { token: 'punctuation.section.scope.elixir',
regex: '\\{',
push:
[ { token: 'punctuation.section.scope.elixir',
regex: '\\}',
next: 'pop' },
{ include: '#nest_curly_and_self' } ] },
{ include: '$self' } ],
'#regex_sub':
[ { include: '#interpolated_elixir' },
{ include: '#escaped_char' },
{ token:
[ 'punctuation.definition.arbitrary-repitition.elixir',
'string.regexp.arbitrary-repitition.elixir',
'string.regexp.arbitrary-repitition.elixir',
'punctuation.definition.arbitrary-repitition.elixir' ],
regex: '(\\{)(\\d+)((?:,\\d+)?)(\\})' },
{ token: 'punctuation.definition.character-class.elixir',
regex: '\\[(?:\\^?\\])?',
push:
[ { token: 'punctuation.definition.character-class.elixir',
regex: '\\]',
next: 'pop' },
{ include: '#escaped_char' },
{ defaultToken: 'string.regexp.character-class.elixir' } ] },
{ token: 'punctuation.definition.group.elixir',
regex: '\\(',
push:
[ { token: 'punctuation.definition.group.elixir',
regex: '\\)',
next: 'pop' },
{ include: '#regex_sub' },
{ defaultToken: 'string.regexp.group.elixir' } ] },
{ token:
[ 'punctuation.definition.comment.elixir',
'comment.line.number-sign.elixir' ],
regex: '(?:^|\\s)(#)(\\s[[a-zA-Z0-9,. \\t?!-][^\\x00-\\x7F]]*$)',
originalRegex: '(?<=^|\\s)(#)\\s[[a-zA-Z0-9,. \\t?!-][^\\x{00}-\\x{7F}]]*$',
comment: 'We are restrictive in what we allow to go after the comment character to avoid false positives, since the availability of comments depend on regexp flags.' } ] };
this.normalizeRules();
};
ElixirHighlightRules.metaData = { comment: 'Textmate bundle for Elixir Programming Language.',
fileTypes: [ 'ex', 'exs' ],
firstLineMatch: '^#!/.*\\belixir',
foldingStartMarker: '(after|else|catch|rescue|\\-\\>|\\{|\\[|do)\\s*$',
foldingStopMarker: '^\\s*((\\}|\\]|after|else|catch|rescue)\\s*$|end\\b)',
keyEquivalent: '^~E',
name: 'Elixir',
scopeName: 'source.elixir' };
oop.inherits(ElixirHighlightRules, TextHighlightRules);
exports.ElixirHighlightRules = ElixirHighlightRules;
});
ace.define("ace/mode/folding/coffee",[], function(require, exports, module) {
"use strict";
var oop = require("../../lib/oop");
var BaseFoldMode = require("./fold_mode").FoldMode;
var Range = require("../../range").Range;
var FoldMode = exports.FoldMode = function() {};
oop.inherits(FoldMode, BaseFoldMode);
(function() {
this.getFoldWidgetRange = function(session, foldStyle, row) {
var range = this.indentationBlock(session, row);
if (range)
return range;
var re = /\S/;
var line = session.getLine(row);
var startLevel = line.search(re);
if (startLevel == -1 || line[startLevel] != "#")
return;
var startColumn = line.length;
var maxRow = session.getLength();
var startRow = row;
var endRow = row;
while (++row < maxRow) {
line = session.getLine(row);
var level = line.search(re);
if (level == -1)
continue;
if (line[level] != "#")
break;
endRow = row;
}
if (endRow > startRow) {
var endColumn = session.getLine(endRow).length;
return new Range(startRow, startColumn, endRow, endColumn);
}
};
this.getFoldWidget = function(session, foldStyle, row) {
var line = session.getLine(row);
var indent = line.search(/\S/);
var next = session.getLine(row + 1);
var prev = session.getLine(row - 1);
var prevIndent = prev.search(/\S/);
var nextIndent = next.search(/\S/);
if (indent == -1) {
session.foldWidgets[row - 1] = prevIndent!= -1 && prevIndent < nextIndent ? "start" : "";
return "";
}
if (prevIndent == -1) {
if (indent == nextIndent && line[indent] == "#" && next[indent] == "#") {
session.foldWidgets[row - 1] = "";
session.foldWidgets[row + 1] = "";
return "start";
}
} else if (prevIndent == indent && line[indent] == "#" && prev[indent] == "#") {
if (session.getLine(row - 2).search(/\S/) == -1) {
session.foldWidgets[row - 1] = "start";
session.foldWidgets[row + 1] = "";
return "";
}
}
if (prevIndent!= -1 && prevIndent < indent)
session.foldWidgets[row - 1] = "start";
else
session.foldWidgets[row - 1] = "";
if (indent < nextIndent)
return "start";
else
return "";
};
}).call(FoldMode.prototype);
});
ace.define("ace/mode/elixir",[], function(require, exports, module) {
"use strict";
var oop = require("../lib/oop");
var TextMode = require("./text").Mode;
var ElixirHighlightRules = require("./elixir_highlight_rules").ElixirHighlightRules;
var FoldMode = require("./folding/coffee").FoldMode;
var Mode = function() {
this.HighlightRules = ElixirHighlightRules;
this.foldingRules = new FoldMode();
this.$behaviour = this.$defaultBehaviour;
};
oop.inherits(Mode, TextMode);
(function() {
this.lineCommentStart = "#";
this.$id = "ace/mode/elixir";
}).call(Mode.prototype);
exports.Mode = Mode;
});
(function() {
ace.require(["ace/mode/elixir"], function(m) {
if (typeof module == "object" && typeof exports == "object" && module) {
module.exports = m;
}
});
})();