mirror of
https://github.com/DataTables/DataTables.git
synced 2025-01-20 13:52:10 +01:00
7bec74c02f
- Additional information and better layout for examples. See the src repo for more information.
2515 lines
75 KiB
JavaScript
2515 lines
75 KiB
JavaScript
// XRegExp 1.5.1
|
|
// (c) 2007-2012 Steven Levithan
|
|
// MIT License
|
|
// <http://xregexp.com>
|
|
// Provides an augmented, extensible, cross-browser implementation of regular expressions,
|
|
// including support for additional syntax, flags, and methods
|
|
|
|
var XRegExp;
|
|
|
|
if (XRegExp) {
|
|
// Avoid running twice, since that would break references to native globals
|
|
throw Error("can't load XRegExp twice in the same frame");
|
|
}
|
|
|
|
// Run within an anonymous function to protect variables and avoid new globals
|
|
(function (undefined) {
|
|
|
|
//---------------------------------
|
|
// Constructor
|
|
//---------------------------------
|
|
|
|
// Accepts a pattern and flags; returns a new, extended `RegExp` object. Differs from a native
|
|
// regular expression in that additional syntax and flags are supported and cross-browser
|
|
// syntax inconsistencies are ameliorated. `XRegExp(/regex/)` clones an existing regex and
|
|
// converts to type XRegExp
|
|
XRegExp = function (pattern, flags) {
|
|
var output = [],
|
|
currScope = XRegExp.OUTSIDE_CLASS,
|
|
pos = 0,
|
|
context, tokenResult, match, chr, regex;
|
|
|
|
if (XRegExp.isRegExp(pattern)) {
|
|
if (flags !== undefined)
|
|
throw TypeError("can't supply flags when constructing one RegExp from another");
|
|
return clone(pattern);
|
|
}
|
|
// Tokens become part of the regex construction process, so protect against infinite
|
|
// recursion when an XRegExp is constructed within a token handler or trigger
|
|
if (isInsideConstructor)
|
|
throw Error("can't call the XRegExp constructor within token definition functions");
|
|
|
|
flags = flags || "";
|
|
context = { // `this` object for custom tokens
|
|
hasNamedCapture: false,
|
|
captureNames: [],
|
|
hasFlag: function (flag) {return flags.indexOf(flag) > -1;},
|
|
setFlag: function (flag) {flags += flag;}
|
|
};
|
|
|
|
while (pos < pattern.length) {
|
|
// Check for custom tokens at the current position
|
|
tokenResult = runTokens(pattern, pos, currScope, context);
|
|
|
|
if (tokenResult) {
|
|
output.push(tokenResult.output);
|
|
pos += (tokenResult.match[0].length || 1);
|
|
} else {
|
|
// Check for native multicharacter metasequences (excluding character classes) at
|
|
// the current position
|
|
if (match = nativ.exec.call(nativeTokens[currScope], pattern.slice(pos))) {
|
|
output.push(match[0]);
|
|
pos += match[0].length;
|
|
} else {
|
|
chr = pattern.charAt(pos);
|
|
if (chr === "[")
|
|
currScope = XRegExp.INSIDE_CLASS;
|
|
else if (chr === "]")
|
|
currScope = XRegExp.OUTSIDE_CLASS;
|
|
// Advance position one character
|
|
output.push(chr);
|
|
pos++;
|
|
}
|
|
}
|
|
}
|
|
|
|
regex = RegExp(output.join(""), nativ.replace.call(flags, flagClip, ""));
|
|
regex._xregexp = {
|
|
source: pattern,
|
|
captureNames: context.hasNamedCapture ? context.captureNames : null
|
|
};
|
|
return regex;
|
|
};
|
|
|
|
|
|
//---------------------------------
|
|
// Public properties
|
|
//---------------------------------
|
|
|
|
XRegExp.version = "1.5.1";
|
|
|
|
// Token scope bitflags
|
|
XRegExp.INSIDE_CLASS = 1;
|
|
XRegExp.OUTSIDE_CLASS = 2;
|
|
|
|
|
|
//---------------------------------
|
|
// Private variables
|
|
//---------------------------------
|
|
|
|
var replacementToken = /\$(?:(\d\d?|[$&`'])|{([$\w]+)})/g,
|
|
flagClip = /[^gimy]+|([\s\S])(?=[\s\S]*\1)/g, // Nonnative and duplicate flags
|
|
quantifier = /^(?:[?*+]|{\d+(?:,\d*)?})\??/,
|
|
isInsideConstructor = false,
|
|
tokens = [],
|
|
// Copy native globals for reference ("native" is an ES3 reserved keyword)
|
|
nativ = {
|
|
exec: RegExp.prototype.exec,
|
|
test: RegExp.prototype.test,
|
|
match: String.prototype.match,
|
|
replace: String.prototype.replace,
|
|
split: String.prototype.split
|
|
},
|
|
compliantExecNpcg = nativ.exec.call(/()??/, "")[1] === undefined, // check `exec` handling of nonparticipating capturing groups
|
|
compliantLastIndexIncrement = function () {
|
|
var x = /^/g;
|
|
nativ.test.call(x, "");
|
|
return !x.lastIndex;
|
|
}(),
|
|
hasNativeY = RegExp.prototype.sticky !== undefined,
|
|
nativeTokens = {};
|
|
|
|
// `nativeTokens` match native multicharacter metasequences only (including deprecated octals,
|
|
// excluding character classes)
|
|
nativeTokens[XRegExp.INSIDE_CLASS] = /^(?:\\(?:[0-3][0-7]{0,2}|[4-7][0-7]?|x[\dA-Fa-f]{2}|u[\dA-Fa-f]{4}|c[A-Za-z]|[\s\S]))/;
|
|
nativeTokens[XRegExp.OUTSIDE_CLASS] = /^(?:\\(?:0(?:[0-3][0-7]{0,2}|[4-7][0-7]?)?|[1-9]\d*|x[\dA-Fa-f]{2}|u[\dA-Fa-f]{4}|c[A-Za-z]|[\s\S])|\(\?[:=!]|[?*+]\?|{\d+(?:,\d*)?}\??)/;
|
|
|
|
|
|
//---------------------------------
|
|
// Public methods
|
|
//---------------------------------
|
|
|
|
// Lets you extend or change XRegExp syntax and create custom flags. This is used internally by
|
|
// the XRegExp library and can be used to create XRegExp plugins. This function is intended for
|
|
// users with advanced knowledge of JavaScript's regular expression syntax and behavior. It can
|
|
// be disabled by `XRegExp.freezeTokens`
|
|
XRegExp.addToken = function (regex, handler, scope, trigger) {
|
|
tokens.push({
|
|
pattern: clone(regex, "g" + (hasNativeY ? "y" : "")),
|
|
handler: handler,
|
|
scope: scope || XRegExp.OUTSIDE_CLASS,
|
|
trigger: trigger || null
|
|
});
|
|
};
|
|
|
|
// Accepts a pattern and flags; returns an extended `RegExp` object. If the pattern and flag
|
|
// combination has previously been cached, the cached copy is returned; otherwise the newly
|
|
// created regex is cached
|
|
XRegExp.cache = function (pattern, flags) {
|
|
var key = pattern + "/" + (flags || "");
|
|
return XRegExp.cache[key] || (XRegExp.cache[key] = XRegExp(pattern, flags));
|
|
};
|
|
|
|
// Accepts a `RegExp` instance; returns a copy with the `/g` flag set. The copy has a fresh
|
|
// `lastIndex` (set to zero). If you want to copy a regex without forcing the `global`
|
|
// property, use `XRegExp(regex)`. Do not use `RegExp(regex)` because it will not preserve
|
|
// special properties required for named capture
|
|
XRegExp.copyAsGlobal = function (regex) {
|
|
return clone(regex, "g");
|
|
};
|
|
|
|
// Accepts a string; returns the string with regex metacharacters escaped. The returned string
|
|
// can safely be used at any point within a regex to match the provided literal string. Escaped
|
|
// characters are [ ] { } ( ) * + ? - . , \ ^ $ | # and whitespace
|
|
XRegExp.escape = function (str) {
|
|
return str.replace(/[-[\]{}()*+?.,\\^$|#\s]/g, "\\$&");
|
|
};
|
|
|
|
// Accepts a string to search, regex to search with, position to start the search within the
|
|
// string (default: 0), and an optional Boolean indicating whether matches must start at-or-
|
|
// after the position or at the specified position only. This function ignores the `lastIndex`
|
|
// of the provided regex in its own handling, but updates the property for compatibility
|
|
XRegExp.execAt = function (str, regex, pos, anchored) {
|
|
var r2 = clone(regex, "g" + ((anchored && hasNativeY) ? "y" : "")),
|
|
match;
|
|
r2.lastIndex = pos = pos || 0;
|
|
match = r2.exec(str); // Run the altered `exec` (required for `lastIndex` fix, etc.)
|
|
if (anchored && match && match.index !== pos)
|
|
match = null;
|
|
if (regex.global)
|
|
regex.lastIndex = match ? r2.lastIndex : 0;
|
|
return match;
|
|
};
|
|
|
|
// Breaks the unrestorable link to XRegExp's private list of tokens, thereby preventing
|
|
// syntax and flag changes. Should be run after XRegExp and any plugins are loaded
|
|
XRegExp.freezeTokens = function () {
|
|
XRegExp.addToken = function () {
|
|
throw Error("can't run addToken after freezeTokens");
|
|
};
|
|
};
|
|
|
|
// Accepts any value; returns a Boolean indicating whether the argument is a `RegExp` object.
|
|
// Note that this is also `true` for regex literals and regexes created by the `XRegExp`
|
|
// constructor. This works correctly for variables created in another frame, when `instanceof`
|
|
// and `constructor` checks would fail to work as intended
|
|
XRegExp.isRegExp = function (o) {
|
|
return Object.prototype.toString.call(o) === "[object RegExp]";
|
|
};
|
|
|
|
// Executes `callback` once per match within `str`. Provides a simpler and cleaner way to
|
|
// iterate over regex matches compared to the traditional approaches of subverting
|
|
// `String.prototype.replace` or repeatedly calling `exec` within a `while` loop
|
|
XRegExp.iterate = function (str, regex, callback, context) {
|
|
var r2 = clone(regex, "g"),
|
|
i = -1, match;
|
|
while (match = r2.exec(str)) { // Run the altered `exec` (required for `lastIndex` fix, etc.)
|
|
if (regex.global)
|
|
regex.lastIndex = r2.lastIndex; // Doing this to follow expectations if `lastIndex` is checked within `callback`
|
|
callback.call(context, match, ++i, str, regex);
|
|
if (r2.lastIndex === match.index)
|
|
r2.lastIndex++;
|
|
}
|
|
if (regex.global)
|
|
regex.lastIndex = 0;
|
|
};
|
|
|
|
// Accepts a string and an array of regexes; returns the result of using each successive regex
|
|
// to search within the matches of the previous regex. The array of regexes can also contain
|
|
// objects with `regex` and `backref` properties, in which case the named or numbered back-
|
|
// references specified are passed forward to the next regex or returned. E.g.:
|
|
// var xregexpImgFileNames = XRegExp.matchChain(html, [
|
|
// {regex: /<img\b([^>]+)>/i, backref: 1}, // <img> tag attributes
|
|
// {regex: XRegExp('(?ix) \\s src=" (?<src> [^"]+ )'), backref: "src"}, // src attribute values
|
|
// {regex: XRegExp("^http://xregexp\\.com(/[^#?]+)", "i"), backref: 1}, // xregexp.com paths
|
|
// /[^\/]+$/ // filenames (strip directory paths)
|
|
// ]);
|
|
XRegExp.matchChain = function (str, chain) {
|
|
return function recurseChain (values, level) {
|
|
var item = chain[level].regex ? chain[level] : {regex: chain[level]},
|
|
regex = clone(item.regex, "g"),
|
|
matches = [], i;
|
|
for (i = 0; i < values.length; i++) {
|
|
XRegExp.iterate(values[i], regex, function (match) {
|
|
matches.push(item.backref ? (match[item.backref] || "") : match[0]);
|
|
});
|
|
}
|
|
return ((level === chain.length - 1) || !matches.length) ?
|
|
matches : recurseChain(matches, level + 1);
|
|
}([str], 0);
|
|
};
|
|
|
|
|
|
//---------------------------------
|
|
// New RegExp prototype methods
|
|
//---------------------------------
|
|
|
|
// Accepts a context object and arguments array; returns the result of calling `exec` with the
|
|
// first value in the arguments array. the context is ignored but is accepted for congruity
|
|
// with `Function.prototype.apply`
|
|
RegExp.prototype.apply = function (context, args) {
|
|
return this.exec(args[0]);
|
|
};
|
|
|
|
// Accepts a context object and string; returns the result of calling `exec` with the provided
|
|
// string. the context is ignored but is accepted for congruity with `Function.prototype.call`
|
|
RegExp.prototype.call = function (context, str) {
|
|
return this.exec(str);
|
|
};
|
|
|
|
|
|
//---------------------------------
|
|
// Overriden native methods
|
|
//---------------------------------
|
|
|
|
// Adds named capture support (with backreferences returned as `result.name`), and fixes two
|
|
// cross-browser issues per ES3:
|
|
// - Captured values for nonparticipating capturing groups should be returned as `undefined`,
|
|
// rather than the empty string.
|
|
// - `lastIndex` should not be incremented after zero-length matches.
|
|
RegExp.prototype.exec = function (str) {
|
|
var match, name, r2, origLastIndex;
|
|
if (!this.global)
|
|
origLastIndex = this.lastIndex;
|
|
match = nativ.exec.apply(this, arguments);
|
|
if (match) {
|
|
// Fix browsers whose `exec` methods don't consistently return `undefined` for
|
|
// nonparticipating capturing groups
|
|
if (!compliantExecNpcg && match.length > 1 && indexOf(match, "") > -1) {
|
|
r2 = RegExp(this.source, nativ.replace.call(getNativeFlags(this), "g", ""));
|
|
// Using `str.slice(match.index)` rather than `match[0]` in case lookahead allowed
|
|
// matching due to characters outside the match
|
|
nativ.replace.call((str + "").slice(match.index), r2, function () {
|
|
for (var i = 1; i < arguments.length - 2; i++) {
|
|
if (arguments[i] === undefined)
|
|
match[i] = undefined;
|
|
}
|
|
});
|
|
}
|
|
// Attach named capture properties
|
|
if (this._xregexp && this._xregexp.captureNames) {
|
|
for (var i = 1; i < match.length; i++) {
|
|
name = this._xregexp.captureNames[i - 1];
|
|
if (name)
|
|
match[name] = match[i];
|
|
}
|
|
}
|
|
// Fix browsers that increment `lastIndex` after zero-length matches
|
|
if (!compliantLastIndexIncrement && this.global && !match[0].length && (this.lastIndex > match.index))
|
|
this.lastIndex--;
|
|
}
|
|
if (!this.global)
|
|
this.lastIndex = origLastIndex; // Fix IE, Opera bug (last tested IE 9.0.5, Opera 11.61 on Windows)
|
|
return match;
|
|
};
|
|
|
|
// Fix browser bugs in native method
|
|
RegExp.prototype.test = function (str) {
|
|
// Use the native `exec` to skip some processing overhead, even though the altered
|
|
// `exec` would take care of the `lastIndex` fixes
|
|
var match, origLastIndex;
|
|
if (!this.global)
|
|
origLastIndex = this.lastIndex;
|
|
match = nativ.exec.call(this, str);
|
|
// Fix browsers that increment `lastIndex` after zero-length matches
|
|
if (match && !compliantLastIndexIncrement && this.global && !match[0].length && (this.lastIndex > match.index))
|
|
this.lastIndex--;
|
|
if (!this.global)
|
|
this.lastIndex = origLastIndex; // Fix IE, Opera bug (last tested IE 9.0.5, Opera 11.61 on Windows)
|
|
return !!match;
|
|
};
|
|
|
|
// Adds named capture support and fixes browser bugs in native method
|
|
String.prototype.match = function (regex) {
|
|
if (!XRegExp.isRegExp(regex))
|
|
regex = RegExp(regex); // Native `RegExp`
|
|
if (regex.global) {
|
|
var result = nativ.match.apply(this, arguments);
|
|
regex.lastIndex = 0; // Fix IE bug
|
|
return result;
|
|
}
|
|
return regex.exec(this); // Run the altered `exec`
|
|
};
|
|
|
|
// Adds support for `${n}` tokens for named and numbered backreferences in replacement text,
|
|
// and provides named backreferences to replacement functions as `arguments[0].name`. Also
|
|
// fixes cross-browser differences in replacement text syntax when performing a replacement
|
|
// using a nonregex search value, and the value of replacement regexes' `lastIndex` property
|
|
// during replacement iterations. Note that this doesn't support SpiderMonkey's proprietary
|
|
// third (`flags`) parameter
|
|
String.prototype.replace = function (search, replacement) {
|
|
var isRegex = XRegExp.isRegExp(search),
|
|
captureNames, result, str, origLastIndex;
|
|
|
|
// There are too many combinations of search/replacement types/values and browser bugs that
|
|
// preclude passing to native `replace`, so don't try
|
|
//if (...)
|
|
// return nativ.replace.apply(this, arguments);
|
|
|
|
if (isRegex) {
|
|
if (search._xregexp)
|
|
captureNames = search._xregexp.captureNames; // Array or `null`
|
|
if (!search.global)
|
|
origLastIndex = search.lastIndex;
|
|
} else {
|
|
search = search + ""; // Type conversion
|
|
}
|
|
|
|
if (Object.prototype.toString.call(replacement) === "[object Function]") {
|
|
result = nativ.replace.call(this + "", search, function () {
|
|
if (captureNames) {
|
|
// Change the `arguments[0]` string primitive to a String object which can store properties
|
|
arguments[0] = new String(arguments[0]);
|
|
// Store named backreferences on `arguments[0]`
|
|
for (var i = 0; i < captureNames.length; i++) {
|
|
if (captureNames[i])
|
|
arguments[0][captureNames[i]] = arguments[i + 1];
|
|
}
|
|
}
|
|
// Update `lastIndex` before calling `replacement` (fix browsers)
|
|
if (isRegex && search.global)
|
|
search.lastIndex = arguments[arguments.length - 2] + arguments[0].length;
|
|
return replacement.apply(null, arguments);
|
|
});
|
|
} else {
|
|
str = this + ""; // Type conversion, so `args[args.length - 1]` will be a string (given nonstring `this`)
|
|
result = nativ.replace.call(str, search, function () {
|
|
var args = arguments; // Keep this function's `arguments` available through closure
|
|
return nativ.replace.call(replacement + "", replacementToken, function ($0, $1, $2) {
|
|
// Numbered backreference (without delimiters) or special variable
|
|
if ($1) {
|
|
switch ($1) {
|
|
case "$": return "$";
|
|
case "&": return args[0];
|
|
case "`": return args[args.length - 1].slice(0, args[args.length - 2]);
|
|
case "'": return args[args.length - 1].slice(args[args.length - 2] + args[0].length);
|
|
// Numbered backreference
|
|
default:
|
|
// What does "$10" mean?
|
|
// - Backreference 10, if 10 or more capturing groups exist
|
|
// - Backreference 1 followed by "0", if 1-9 capturing groups exist
|
|
// - Otherwise, it's the string "$10"
|
|
// Also note:
|
|
// - Backreferences cannot be more than two digits (enforced by `replacementToken`)
|
|
// - "$01" is equivalent to "$1" if a capturing group exists, otherwise it's the string "$01"
|
|
// - There is no "$0" token ("$&" is the entire match)
|
|
var literalNumbers = "";
|
|
$1 = +$1; // Type conversion; drop leading zero
|
|
if (!$1) // `$1` was "0" or "00"
|
|
return $0;
|
|
while ($1 > args.length - 3) {
|
|
literalNumbers = String.prototype.slice.call($1, -1) + literalNumbers;
|
|
$1 = Math.floor($1 / 10); // Drop the last digit
|
|
}
|
|
return ($1 ? args[$1] || "" : "$") + literalNumbers;
|
|
}
|
|
// Named backreference or delimited numbered backreference
|
|
} else {
|
|
// What does "${n}" mean?
|
|
// - Backreference to numbered capture n. Two differences from "$n":
|
|
// - n can be more than two digits
|
|
// - Backreference 0 is allowed, and is the entire match
|
|
// - Backreference to named capture n, if it exists and is not a number overridden by numbered capture
|
|
// - Otherwise, it's the string "${n}"
|
|
var n = +$2; // Type conversion; drop leading zeros
|
|
if (n <= args.length - 3)
|
|
return args[n];
|
|
n = captureNames ? indexOf(captureNames, $2) : -1;
|
|
return n > -1 ? args[n + 1] : $0;
|
|
}
|
|
});
|
|
});
|
|
}
|
|
|
|
if (isRegex) {
|
|
if (search.global)
|
|
search.lastIndex = 0; // Fix IE, Safari bug (last tested IE 9.0.5, Safari 5.1.2 on Windows)
|
|
else
|
|
search.lastIndex = origLastIndex; // Fix IE, Opera bug (last tested IE 9.0.5, Opera 11.61 on Windows)
|
|
}
|
|
|
|
return result;
|
|
};
|
|
|
|
// A consistent cross-browser, ES3 compliant `split`
|
|
String.prototype.split = function (s /* separator */, limit) {
|
|
// If separator `s` is not a regex, use the native `split`
|
|
if (!XRegExp.isRegExp(s))
|
|
return nativ.split.apply(this, arguments);
|
|
|
|
var str = this + "", // Type conversion
|
|
output = [],
|
|
lastLastIndex = 0,
|
|
match, lastLength;
|
|
|
|
// Behavior for `limit`: if it's...
|
|
// - `undefined`: No limit
|
|
// - `NaN` or zero: Return an empty array
|
|
// - A positive number: Use `Math.floor(limit)`
|
|
// - A negative number: No limit
|
|
// - Other: Type-convert, then use the above rules
|
|
if (limit === undefined || +limit < 0) {
|
|
limit = Infinity;
|
|
} else {
|
|
limit = Math.floor(+limit);
|
|
if (!limit)
|
|
return [];
|
|
}
|
|
|
|
// This is required if not `s.global`, and it avoids needing to set `s.lastIndex` to zero
|
|
// and restore it to its original value when we're done using the regex
|
|
s = XRegExp.copyAsGlobal(s);
|
|
|
|
while (match = s.exec(str)) { // Run the altered `exec` (required for `lastIndex` fix, etc.)
|
|
if (s.lastIndex > lastLastIndex) {
|
|
output.push(str.slice(lastLastIndex, match.index));
|
|
|
|
if (match.length > 1 && match.index < str.length)
|
|
Array.prototype.push.apply(output, match.slice(1));
|
|
|
|
lastLength = match[0].length;
|
|
lastLastIndex = s.lastIndex;
|
|
|
|
if (output.length >= limit)
|
|
break;
|
|
}
|
|
|
|
if (s.lastIndex === match.index)
|
|
s.lastIndex++;
|
|
}
|
|
|
|
if (lastLastIndex === str.length) {
|
|
if (!nativ.test.call(s, "") || lastLength)
|
|
output.push("");
|
|
} else {
|
|
output.push(str.slice(lastLastIndex));
|
|
}
|
|
|
|
return output.length > limit ? output.slice(0, limit) : output;
|
|
};
|
|
|
|
|
|
//---------------------------------
|
|
// Private helper functions
|
|
//---------------------------------
|
|
|
|
// Supporting function for `XRegExp`, `XRegExp.copyAsGlobal`, etc. Returns a copy of a `RegExp`
|
|
// instance with a fresh `lastIndex` (set to zero), preserving properties required for named
|
|
// capture. Also allows adding new flags in the process of copying the regex
|
|
function clone (regex, additionalFlags) {
|
|
if (!XRegExp.isRegExp(regex))
|
|
throw TypeError("type RegExp expected");
|
|
var x = regex._xregexp;
|
|
regex = XRegExp(regex.source, getNativeFlags(regex) + (additionalFlags || ""));
|
|
if (x) {
|
|
regex._xregexp = {
|
|
source: x.source,
|
|
captureNames: x.captureNames ? x.captureNames.slice(0) : null
|
|
};
|
|
}
|
|
return regex;
|
|
}
|
|
|
|
function getNativeFlags (regex) {
|
|
return (regex.global ? "g" : "") +
|
|
(regex.ignoreCase ? "i" : "") +
|
|
(regex.multiline ? "m" : "") +
|
|
(regex.extended ? "x" : "") + // Proposed for ES4; included in AS3
|
|
(regex.sticky ? "y" : "");
|
|
}
|
|
|
|
function runTokens (pattern, index, scope, context) {
|
|
var i = tokens.length,
|
|
result, match, t;
|
|
// Protect against constructing XRegExps within token handler and trigger functions
|
|
isInsideConstructor = true;
|
|
// Must reset `isInsideConstructor`, even if a `trigger` or `handler` throws
|
|
try {
|
|
while (i--) { // Run in reverse order
|
|
t = tokens[i];
|
|
if ((scope & t.scope) && (!t.trigger || t.trigger.call(context))) {
|
|
t.pattern.lastIndex = index;
|
|
match = t.pattern.exec(pattern); // Running the altered `exec` here allows use of named backreferences, etc.
|
|
if (match && match.index === index) {
|
|
result = {
|
|
output: t.handler.call(context, match, scope),
|
|
match: match
|
|
};
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
} catch (err) {
|
|
throw err;
|
|
} finally {
|
|
isInsideConstructor = false;
|
|
}
|
|
return result;
|
|
}
|
|
|
|
function indexOf (array, item, from) {
|
|
if (Array.prototype.indexOf) // Use the native array method if available
|
|
return array.indexOf(item, from);
|
|
for (var i = from || 0; i < array.length; i++) {
|
|
if (array[i] === item)
|
|
return i;
|
|
}
|
|
return -1;
|
|
}
|
|
|
|
|
|
//---------------------------------
|
|
// Built-in tokens
|
|
//---------------------------------
|
|
|
|
// Augment XRegExp's regular expression syntax and flags. Note that when adding tokens, the
|
|
// third (`scope`) argument defaults to `XRegExp.OUTSIDE_CLASS`
|
|
|
|
// Comment pattern: (?# )
|
|
XRegExp.addToken(
|
|
/\(\?#[^)]*\)/,
|
|
function (match) {
|
|
// Keep tokens separated unless the following token is a quantifier
|
|
return nativ.test.call(quantifier, match.input.slice(match.index + match[0].length)) ? "" : "(?:)";
|
|
}
|
|
);
|
|
|
|
// Capturing group (match the opening parenthesis only).
|
|
// Required for support of named capturing groups
|
|
XRegExp.addToken(
|
|
/\((?!\?)/,
|
|
function () {
|
|
this.captureNames.push(null);
|
|
return "(";
|
|
}
|
|
);
|
|
|
|
// Named capturing group (match the opening delimiter only): (?<name>
|
|
XRegExp.addToken(
|
|
/\(\?<([$\w]+)>/,
|
|
function (match) {
|
|
this.captureNames.push(match[1]);
|
|
this.hasNamedCapture = true;
|
|
return "(";
|
|
}
|
|
);
|
|
|
|
// Named backreference: \k<name>
|
|
XRegExp.addToken(
|
|
/\\k<([\w$]+)>/,
|
|
function (match) {
|
|
var index = indexOf(this.captureNames, match[1]);
|
|
// Keep backreferences separate from subsequent literal numbers. Preserve back-
|
|
// references to named groups that are undefined at this point as literal strings
|
|
return index > -1 ?
|
|
"\\" + (index + 1) + (isNaN(match.input.charAt(match.index + match[0].length)) ? "" : "(?:)") :
|
|
match[0];
|
|
}
|
|
);
|
|
|
|
// Empty character class: [] or [^]
|
|
XRegExp.addToken(
|
|
/\[\^?]/,
|
|
function (match) {
|
|
// For cross-browser compatibility with ES3, convert [] to \b\B and [^] to [\s\S].
|
|
// (?!) should work like \b\B, but is unreliable in Firefox
|
|
return match[0] === "[]" ? "\\b\\B" : "[\\s\\S]";
|
|
}
|
|
);
|
|
|
|
// Mode modifier at the start of the pattern only, with any combination of flags imsx: (?imsx)
|
|
// Does not support x(?i), (?-i), (?i-m), (?i: ), (?i)(?m), etc.
|
|
XRegExp.addToken(
|
|
/^\(\?([imsx]+)\)/,
|
|
function (match) {
|
|
this.setFlag(match[1]);
|
|
return "";
|
|
}
|
|
);
|
|
|
|
// Whitespace and comments, in free-spacing (aka extended) mode only
|
|
XRegExp.addToken(
|
|
/(?:\s+|#.*)+/,
|
|
function (match) {
|
|
// Keep tokens separated unless the following token is a quantifier
|
|
return nativ.test.call(quantifier, match.input.slice(match.index + match[0].length)) ? "" : "(?:)";
|
|
},
|
|
XRegExp.OUTSIDE_CLASS,
|
|
function () {return this.hasFlag("x");}
|
|
);
|
|
|
|
// Dot, in dotall (aka singleline) mode only
|
|
XRegExp.addToken(
|
|
/\./,
|
|
function () {return "[\\s\\S]";},
|
|
XRegExp.OUTSIDE_CLASS,
|
|
function () {return this.hasFlag("s");}
|
|
);
|
|
|
|
|
|
//---------------------------------
|
|
// Backward compatibility
|
|
//---------------------------------
|
|
|
|
// Uncomment the following block for compatibility with XRegExp 1.0-1.2:
|
|
/*
|
|
XRegExp.matchWithinChain = XRegExp.matchChain;
|
|
RegExp.prototype.addFlags = function (s) {return clone(this, s);};
|
|
RegExp.prototype.execAll = function (s) {var r = []; XRegExp.iterate(s, this, function (m) {r.push(m);}); return r;};
|
|
RegExp.prototype.forEachExec = function (s, f, c) {return XRegExp.iterate(s, this, f, c);};
|
|
RegExp.prototype.validate = function (s) {var r = RegExp("^(?:" + this.source + ")$(?!\\s)", getNativeFlags(this)); if (this.global) this.lastIndex = 0; return s.search(r) === 0;};
|
|
*/
|
|
|
|
})();
|
|
|
|
//
|
|
// Begin anonymous function. This is used to contain local scope variables without polutting global scope.
|
|
//
|
|
if (typeof(SyntaxHighlighter) == 'undefined') var SyntaxHighlighter = function() {
|
|
|
|
// CommonJS
|
|
if (typeof(require) != 'undefined' && typeof(XRegExp) == 'undefined')
|
|
{
|
|
XRegExp = require('XRegExp').XRegExp;
|
|
}
|
|
|
|
// Shortcut object which will be assigned to the SyntaxHighlighter variable.
|
|
// This is a shorthand for local reference in order to avoid long namespace
|
|
// references to SyntaxHighlighter.whatever...
|
|
var sh = {
|
|
defaults : {
|
|
/** Additional CSS class names to be added to highlighter elements. */
|
|
'class-name' : '',
|
|
|
|
/** First line number. */
|
|
'first-line' : 1,
|
|
|
|
/**
|
|
* Pads line numbers. Possible values are:
|
|
*
|
|
* false - don't pad line numbers.
|
|
* true - automaticaly pad numbers with minimum required number of leading zeroes.
|
|
* [int] - length up to which pad line numbers.
|
|
*/
|
|
'pad-line-numbers' : false,
|
|
|
|
/** Lines to highlight. */
|
|
'highlight' : null,
|
|
|
|
/** Title to be displayed above the code block. */
|
|
'title' : null,
|
|
|
|
/** Enables or disables smart tabs. */
|
|
'smart-tabs' : true,
|
|
|
|
/** Gets or sets tab size. */
|
|
'tab-size' : 4,
|
|
|
|
/** Enables or disables gutter. */
|
|
'gutter' : true,
|
|
|
|
/** Enables or disables toolbar. */
|
|
'toolbar' : true,
|
|
|
|
/** Enables quick code copy and paste from double click. */
|
|
'quick-code' : true,
|
|
|
|
/** Forces code view to be collapsed. */
|
|
'collapse' : false,
|
|
|
|
/** Enables or disables automatic links. */
|
|
'auto-links' : true,
|
|
|
|
/** Gets or sets light mode. Equavalent to turning off gutter and toolbar. */
|
|
'light' : false,
|
|
|
|
'unindent' : true,
|
|
|
|
'html-script' : false
|
|
},
|
|
|
|
config : {
|
|
space : ' ',
|
|
|
|
/** Enables use of <SCRIPT type="syntaxhighlighter" /> tags. */
|
|
useScriptTags : true,
|
|
|
|
/** Blogger mode flag. */
|
|
bloggerMode : false,
|
|
|
|
stripBrs : false,
|
|
|
|
/** Name of the tag that SyntaxHighlighter will automatically look for. */
|
|
tagName : 'pre',
|
|
|
|
strings : {
|
|
expandSource : 'expand source',
|
|
help : '?',
|
|
alert: 'SyntaxHighlighter\n\n',
|
|
noBrush : 'Can\'t find brush for: ',
|
|
brushNotHtmlScript : 'Brush wasn\'t configured for html-script option: ',
|
|
|
|
// this is populated by the build script
|
|
aboutDialog : '@ABOUT@'
|
|
}
|
|
},
|
|
|
|
/** Internal 'global' variables. */
|
|
vars : {
|
|
discoveredBrushes : null,
|
|
highlighters : {}
|
|
},
|
|
|
|
/** This object is populated by user included external brush files. */
|
|
brushes : {},
|
|
|
|
/** Common regular expressions. */
|
|
regexLib : {
|
|
multiLineCComments : /\/\*[\s\S]*?\*\//gm,
|
|
singleLineCComments : /\/\/.*$/gm,
|
|
singleLinePerlComments : /#.*$/gm,
|
|
doubleQuotedString : /"([^\\"\n]|\\.)*"/g,
|
|
singleQuotedString : /'([^\\'\n]|\\.)*'/g,
|
|
multiLineDoubleQuotedString : new XRegExp('"([^\\\\"]|\\\\.)*"', 'gs'),
|
|
multiLineSingleQuotedString : new XRegExp("'([^\\\\']|\\\\.)*'", 'gs'),
|
|
xmlComments : /(<|<)!--[\s\S]*?--(>|>)/gm,
|
|
url : /\w+:\/\/[\w-.\/?%&=:@;#]*/g,
|
|
|
|
/** <?= ?> tags. */
|
|
phpScriptTags : { left: /(<|<)\?(?:=|php)?/g, right: /\?(>|>)/g, 'eof' : true },
|
|
|
|
/** <%= %> tags. */
|
|
aspScriptTags : { left: /(<|<)%=?/g, right: /%(>|>)/g },
|
|
|
|
/** <script> tags. */
|
|
scriptScriptTags : { left: /(<|<)\s*script.*?(>|>)/gi, right: /(<|<)\/\s*script\s*(>|>)/gi }
|
|
},
|
|
|
|
toolbar: {
|
|
/**
|
|
* Generates HTML markup for the toolbar.
|
|
* @param {Highlighter} highlighter Highlighter instance.
|
|
* @return {String} Returns HTML markup.
|
|
*/
|
|
getHtml: function(highlighter)
|
|
{
|
|
var html = '<div class="toolbar">',
|
|
items = sh.toolbar.items,
|
|
list = items.list
|
|
;
|
|
|
|
function defaultGetHtml(highlighter, name)
|
|
{
|
|
return sh.toolbar.getButtonHtml(highlighter, name, sh.config.strings[name]);
|
|
};
|
|
|
|
for (var i = 0; i < list.length; i++)
|
|
html += (items[list[i]].getHtml || defaultGetHtml)(highlighter, list[i]);
|
|
|
|
html += '</div>';
|
|
|
|
return html;
|
|
},
|
|
|
|
/**
|
|
* Generates HTML markup for a regular button in the toolbar.
|
|
* @param {Highlighter} highlighter Highlighter instance.
|
|
* @param {String} commandName Command name that would be executed.
|
|
* @param {String} label Label text to display.
|
|
* @return {String} Returns HTML markup.
|
|
*/
|
|
getButtonHtml: function(highlighter, commandName, label)
|
|
{
|
|
return '<span><a href="#" class="toolbar_item'
|
|
+ ' command_' + commandName
|
|
+ ' ' + commandName
|
|
+ '">' + label + '</a></span>'
|
|
;
|
|
},
|
|
|
|
/**
|
|
* Event handler for a toolbar anchor.
|
|
*/
|
|
handler: function(e)
|
|
{
|
|
var target = e.target,
|
|
className = target.className || ''
|
|
;
|
|
|
|
function getValue(name)
|
|
{
|
|
var r = new RegExp(name + '_(\\w+)'),
|
|
match = r.exec(className)
|
|
;
|
|
|
|
return match ? match[1] : null;
|
|
};
|
|
|
|
var highlighter = getHighlighterById(findParentElement(target, '.syntaxhighlighter').id),
|
|
commandName = getValue('command')
|
|
;
|
|
|
|
// execute the toolbar command
|
|
if (highlighter && commandName)
|
|
sh.toolbar.items[commandName].execute(highlighter);
|
|
|
|
// disable default A click behaviour
|
|
e.preventDefault();
|
|
},
|
|
|
|
/** Collection of toolbar items. */
|
|
items : {
|
|
// Ordered lis of items in the toolbar. Can't expect `for (var n in items)` to be consistent.
|
|
list: ['expandSource', 'help'],
|
|
|
|
expandSource: {
|
|
getHtml: function(highlighter)
|
|
{
|
|
if (highlighter.getParam('collapse') != true)
|
|
return '';
|
|
|
|
var title = highlighter.getParam('title');
|
|
return sh.toolbar.getButtonHtml(highlighter, 'expandSource', title ? title : sh.config.strings.expandSource);
|
|
},
|
|
|
|
execute: function(highlighter)
|
|
{
|
|
var div = getHighlighterDivById(highlighter.id);
|
|
removeClass(div, 'collapsed');
|
|
}
|
|
},
|
|
|
|
/** Command to display the about dialog window. */
|
|
help: {
|
|
execute: function(highlighter)
|
|
{
|
|
var wnd = popup('', '_blank', 500, 250, 'scrollbars=0'),
|
|
doc = wnd.document
|
|
;
|
|
|
|
doc.write(sh.config.strings.aboutDialog);
|
|
doc.close();
|
|
wnd.focus();
|
|
}
|
|
}
|
|
}
|
|
},
|
|
|
|
/**
|
|
* Finds all elements on the page which should be processes by SyntaxHighlighter.
|
|
*
|
|
* @param {Object} globalParams Optional parameters which override element's
|
|
* parameters. Only used if element is specified.
|
|
*
|
|
* @param {Object} element Optional element to highlight. If none is
|
|
* provided, all elements in the current document
|
|
* are returned which qualify.
|
|
*
|
|
* @return {Array} Returns list of <code>{ target: DOMElement, params: Object }</code> objects.
|
|
*/
|
|
findElements: function(globalParams, element)
|
|
{
|
|
var elements = element ? [element] : toArray(document.getElementsByTagName(sh.config.tagName)),
|
|
conf = sh.config,
|
|
result = []
|
|
;
|
|
|
|
// support for <SCRIPT TYPE="syntaxhighlighter" /> feature
|
|
if (conf.useScriptTags)
|
|
elements = elements.concat(getSyntaxHighlighterScriptTags());
|
|
|
|
if (elements.length === 0)
|
|
return result;
|
|
|
|
for (var i = 0; i < elements.length; i++)
|
|
{
|
|
var item = {
|
|
target: elements[i],
|
|
// local params take precedence over globals
|
|
params: merge(globalParams, parseParams(elements[i].className))
|
|
};
|
|
|
|
if (item.params['brush'] == null)
|
|
continue;
|
|
|
|
result.push(item);
|
|
}
|
|
|
|
return result;
|
|
},
|
|
|
|
/**
|
|
* Shorthand to highlight all elements on the page that are marked as
|
|
* SyntaxHighlighter source code.
|
|
*
|
|
* @param {Object} globalParams Optional parameters which override element's
|
|
* parameters. Only used if element is specified.
|
|
*
|
|
* @param {Object} element Optional element to highlight. If none is
|
|
* provided, all elements in the current document
|
|
* are highlighted.
|
|
*/
|
|
highlight: function(globalParams, element)
|
|
{
|
|
var elements = this.findElements(globalParams, element),
|
|
propertyName = 'innerHTML',
|
|
highlighter = null,
|
|
conf = sh.config
|
|
;
|
|
|
|
if (elements.length === 0)
|
|
return;
|
|
|
|
for (var i = 0; i < elements.length; i++)
|
|
{
|
|
var element = elements[i],
|
|
target = element.target,
|
|
params = element.params,
|
|
brushName = params.brush,
|
|
code
|
|
;
|
|
|
|
if (brushName == null)
|
|
continue;
|
|
|
|
// Instantiate a brush
|
|
if (params['html-script'] == 'true' || sh.defaults['html-script'] == true)
|
|
{
|
|
highlighter = new sh.HtmlScript(brushName);
|
|
brushName = 'htmlscript';
|
|
}
|
|
else
|
|
{
|
|
var brush = findBrush(brushName);
|
|
|
|
if (brush)
|
|
highlighter = new brush();
|
|
else
|
|
continue;
|
|
}
|
|
|
|
code = target[propertyName];
|
|
|
|
// remove CDATA from <SCRIPT/> tags if it's present
|
|
if (conf.useScriptTags)
|
|
code = stripCData(code);
|
|
|
|
// Inject title if the attribute is present
|
|
if ((target.title || '') != '')
|
|
params.title = target.title;
|
|
|
|
params['brush'] = brushName;
|
|
highlighter.init(params);
|
|
element = highlighter.getDiv(code);
|
|
|
|
// carry over ID
|
|
if ((target.id || '') != '')
|
|
element.id = target.id;
|
|
|
|
target.parentNode.replaceChild(element, target);
|
|
}
|
|
},
|
|
|
|
/**
|
|
* Main entry point for the SyntaxHighlighter.
|
|
* @param {Object} params Optional params to apply to all highlighted elements.
|
|
*/
|
|
all: function(params)
|
|
{
|
|
attachEvent(
|
|
window,
|
|
'load',
|
|
function() { sh.highlight(params); }
|
|
);
|
|
}
|
|
}; // end of sh
|
|
|
|
/**
|
|
* Checks if target DOM elements has specified CSS class.
|
|
* @param {DOMElement} target Target DOM element to check.
|
|
* @param {String} className Name of the CSS class to check for.
|
|
* @return {Boolean} Returns true if class name is present, false otherwise.
|
|
*/
|
|
function hasClass(target, className)
|
|
{
|
|
return target.className.indexOf(className) != -1;
|
|
};
|
|
|
|
/**
|
|
* Adds CSS class name to the target DOM element.
|
|
* @param {DOMElement} target Target DOM element.
|
|
* @param {String} className New CSS class to add.
|
|
*/
|
|
function addClass(target, className)
|
|
{
|
|
if (!hasClass(target, className))
|
|
target.className += ' ' + className;
|
|
};
|
|
|
|
/**
|
|
* Removes CSS class name from the target DOM element.
|
|
* @param {DOMElement} target Target DOM element.
|
|
* @param {String} className CSS class to remove.
|
|
*/
|
|
function removeClass(target, className)
|
|
{
|
|
target.className = target.className.replace(className, '');
|
|
};
|
|
|
|
/**
|
|
* Converts the source to array object. Mostly used for function arguments and
|
|
* lists returned by getElementsByTagName() which aren't Array objects.
|
|
* @param {List} source Source list.
|
|
* @return {Array} Returns array.
|
|
*/
|
|
function toArray(source)
|
|
{
|
|
var result = [];
|
|
|
|
for (var i = 0; i < source.length; i++)
|
|
result.push(source[i]);
|
|
|
|
return result;
|
|
};
|
|
|
|
/**
|
|
* Splits block of text into lines.
|
|
* @param {String} block Block of text.
|
|
* @return {Array} Returns array of lines.
|
|
*/
|
|
function splitLines(block)
|
|
{
|
|
return block.split(/\r?\n/);
|
|
}
|
|
|
|
/**
|
|
* Generates HTML ID for the highlighter.
|
|
* @param {String} highlighterId Highlighter ID.
|
|
* @return {String} Returns HTML ID.
|
|
*/
|
|
function getHighlighterId(id)
|
|
{
|
|
var prefix = 'highlighter_';
|
|
return id.indexOf(prefix) == 0 ? id : prefix + id;
|
|
};
|
|
|
|
/**
|
|
* Finds Highlighter instance by ID.
|
|
* @param {String} highlighterId Highlighter ID.
|
|
* @return {Highlighter} Returns instance of the highlighter.
|
|
*/
|
|
function getHighlighterById(id)
|
|
{
|
|
return sh.vars.highlighters[getHighlighterId(id)];
|
|
};
|
|
|
|
/**
|
|
* Finds highlighter's DIV container.
|
|
* @param {String} highlighterId Highlighter ID.
|
|
* @return {Element} Returns highlighter's DIV element.
|
|
*/
|
|
function getHighlighterDivById(id)
|
|
{
|
|
return document.getElementById(getHighlighterId(id));
|
|
};
|
|
|
|
/**
|
|
* Stores highlighter so that getHighlighterById() can do its thing. Each
|
|
* highlighter must call this method to preserve itself.
|
|
* @param {Highilghter} highlighter Highlighter instance.
|
|
*/
|
|
function storeHighlighter(highlighter)
|
|
{
|
|
sh.vars.highlighters[getHighlighterId(highlighter.id)] = highlighter;
|
|
};
|
|
|
|
/**
|
|
* Looks for a child or parent node which has specified classname.
|
|
* Equivalent to jQuery's $(container).find(".className")
|
|
* @param {Element} target Target element.
|
|
* @param {String} search Class name or node name to look for.
|
|
* @param {Boolean} reverse If set to true, will go up the node tree instead of down.
|
|
* @return {Element} Returns found child or parent element on null.
|
|
*/
|
|
function findElement(target, search, reverse /* optional */)
|
|
{
|
|
if (target == null)
|
|
return null;
|
|
|
|
var nodes = reverse != true ? target.childNodes : [ target.parentNode ],
|
|
propertyToFind = { '#' : 'id', '.' : 'className' }[search.substr(0, 1)] || 'nodeName',
|
|
expectedValue,
|
|
found
|
|
;
|
|
|
|
expectedValue = propertyToFind != 'nodeName'
|
|
? search.substr(1)
|
|
: search.toUpperCase()
|
|
;
|
|
|
|
// main return of the found node
|
|
if ((target[propertyToFind] || '').indexOf(expectedValue) != -1)
|
|
return target;
|
|
|
|
for (var i = 0; nodes && i < nodes.length && found == null; i++)
|
|
found = findElement(nodes[i], search, reverse);
|
|
|
|
return found;
|
|
};
|
|
|
|
/**
|
|
* Looks for a parent node which has specified classname.
|
|
* This is an alias to <code>findElement(container, className, true)</code>.
|
|
* @param {Element} target Target element.
|
|
* @param {String} className Class name to look for.
|
|
* @return {Element} Returns found parent element on null.
|
|
*/
|
|
function findParentElement(target, className)
|
|
{
|
|
return findElement(target, className, true);
|
|
};
|
|
|
|
/**
|
|
* Finds an index of element in the array.
|
|
* @ignore
|
|
* @param {Object} searchElement
|
|
* @param {Number} fromIndex
|
|
* @return {Number} Returns index of element if found; -1 otherwise.
|
|
*/
|
|
function indexOf(array, searchElement, fromIndex)
|
|
{
|
|
fromIndex = Math.max(fromIndex || 0, 0);
|
|
|
|
for (var i = fromIndex; i < array.length; i++)
|
|
if(array[i] == searchElement)
|
|
return i;
|
|
|
|
return -1;
|
|
};
|
|
|
|
/**
|
|
* Generates a unique element ID.
|
|
*/
|
|
function guid(prefix)
|
|
{
|
|
return (prefix || '') + Math.round(Math.random() * 1000000).toString();
|
|
};
|
|
|
|
/**
|
|
* Merges two objects. Values from obj2 override values in obj1.
|
|
* Function is NOT recursive and works only for one dimensional objects.
|
|
* @param {Object} obj1 First object.
|
|
* @param {Object} obj2 Second object.
|
|
* @return {Object} Returns combination of both objects.
|
|
*/
|
|
function merge(obj1, obj2)
|
|
{
|
|
var result = {}, name;
|
|
|
|
for (name in obj1)
|
|
result[name] = obj1[name];
|
|
|
|
for (name in obj2)
|
|
result[name] = obj2[name];
|
|
|
|
return result;
|
|
};
|
|
|
|
/**
|
|
* Attempts to convert string to boolean.
|
|
* @param {String} value Input string.
|
|
* @return {Boolean} Returns true if input was "true", false if input was "false" and value otherwise.
|
|
*/
|
|
function toBoolean(value)
|
|
{
|
|
var result = { "true" : true, "false" : false }[value];
|
|
return result == null ? value : result;
|
|
};
|
|
|
|
/**
|
|
* Opens up a centered popup window.
|
|
* @param {String} url URL to open in the window.
|
|
* @param {String} name Popup name.
|
|
* @param {int} width Popup width.
|
|
* @param {int} height Popup height.
|
|
* @param {String} options window.open() options.
|
|
* @return {Window} Returns window instance.
|
|
*/
|
|
function popup(url, name, width, height, options)
|
|
{
|
|
var x = (screen.width - width) / 2,
|
|
y = (screen.height - height) / 2
|
|
;
|
|
|
|
options += ', left=' + x +
|
|
', top=' + y +
|
|
', width=' + width +
|
|
', height=' + height
|
|
;
|
|
options = options.replace(/^,/, '');
|
|
|
|
var win = window.open(url, name, options);
|
|
win.focus();
|
|
return win;
|
|
};
|
|
|
|
/**
|
|
* Adds event handler to the target object.
|
|
* @param {Object} obj Target object.
|
|
* @param {String} type Name of the event.
|
|
* @param {Function} func Handling function.
|
|
*/
|
|
function attachEvent(obj, type, func, scope)
|
|
{
|
|
function handler(e)
|
|
{
|
|
e = e || window.event;
|
|
|
|
if (!e.target)
|
|
{
|
|
e.target = e.srcElement;
|
|
e.preventDefault = function()
|
|
{
|
|
this.returnValue = false;
|
|
};
|
|
}
|
|
|
|
func.call(scope || window, e);
|
|
};
|
|
|
|
if (obj.attachEvent)
|
|
{
|
|
obj.attachEvent('on' + type, handler);
|
|
}
|
|
else
|
|
{
|
|
obj.addEventListener(type, handler, false);
|
|
}
|
|
};
|
|
|
|
/**
|
|
* Displays an alert.
|
|
* @param {String} str String to display.
|
|
*/
|
|
function alert(str)
|
|
{
|
|
window.alert(sh.config.strings.alert + str);
|
|
};
|
|
|
|
/**
|
|
* Finds a brush by its alias.
|
|
*
|
|
* @param {String} alias Brush alias.
|
|
* @param {Boolean} showAlert Suppresses the alert if false.
|
|
* @return {Brush} Returns bursh constructor if found, null otherwise.
|
|
*/
|
|
function findBrush(alias, showAlert)
|
|
{
|
|
var brushes = sh.vars.discoveredBrushes,
|
|
result = null
|
|
;
|
|
|
|
if (brushes == null)
|
|
{
|
|
brushes = {};
|
|
|
|
// Find all brushes
|
|
for (var brush in sh.brushes)
|
|
{
|
|
var info = sh.brushes[brush],
|
|
aliases = info.aliases
|
|
;
|
|
|
|
if (aliases == null)
|
|
continue;
|
|
|
|
// keep the brush name
|
|
info.brushName = brush.toLowerCase();
|
|
|
|
for (var i = 0; i < aliases.length; i++)
|
|
brushes[aliases[i]] = brush;
|
|
}
|
|
|
|
sh.vars.discoveredBrushes = brushes;
|
|
}
|
|
|
|
result = sh.brushes[brushes[alias]];
|
|
|
|
if (result == null && showAlert)
|
|
alert(sh.config.strings.noBrush + alias);
|
|
|
|
return result;
|
|
};
|
|
|
|
/**
|
|
* Executes a callback on each line and replaces each line with result from the callback.
|
|
* @param {Object} str Input string.
|
|
* @param {Object} callback Callback function taking one string argument and returning a string.
|
|
*/
|
|
function eachLine(str, callback)
|
|
{
|
|
var lines = splitLines(str);
|
|
|
|
for (var i = 0; i < lines.length; i++)
|
|
lines[i] = callback(lines[i], i);
|
|
|
|
// include \r to enable copy-paste on windows (ie8) without getting everything on one line
|
|
return lines.join('\r\n');
|
|
};
|
|
|
|
/**
|
|
* This is a special trim which only removes first and last empty lines
|
|
* and doesn't affect valid leading space on the first line.
|
|
*
|
|
* @param {String} str Input string
|
|
* @return {String} Returns string without empty first and last lines.
|
|
*/
|
|
function trimFirstAndLastLines(str)
|
|
{
|
|
return str.replace(/^[ ]*[\n]+|[\n]*[ ]*$/g, '');
|
|
};
|
|
|
|
/**
|
|
* Parses key/value pairs into hash object.
|
|
*
|
|
* Understands the following formats:
|
|
* - name: word;
|
|
* - name: [word, word];
|
|
* - name: "string";
|
|
* - name: 'string';
|
|
*
|
|
* For example:
|
|
* name1: value; name2: [value, value]; name3: 'value'
|
|
*
|
|
* @param {String} str Input string.
|
|
* @return {Object} Returns deserialized object.
|
|
*/
|
|
function parseParams(str)
|
|
{
|
|
var match,
|
|
result = {},
|
|
arrayRegex = new XRegExp("^\\[(?<values>(.*?))\\]$"),
|
|
regex = new XRegExp(
|
|
"(?<name>[\\w-]+)" +
|
|
"\\s*:\\s*" +
|
|
"(?<value>" +
|
|
"[\\w-%#]+|" + // word
|
|
"\\[.*?\\]|" + // [] array
|
|
'".*?"|' + // "" string
|
|
"'.*?'" + // '' string
|
|
")\\s*;?",
|
|
"g"
|
|
)
|
|
;
|
|
|
|
while ((match = regex.exec(str)) != null)
|
|
{
|
|
var value = match.value
|
|
.replace(/^['"]|['"]$/g, '') // strip quotes from end of strings
|
|
;
|
|
|
|
// try to parse array value
|
|
if (value != null && arrayRegex.test(value))
|
|
{
|
|
var m = arrayRegex.exec(value);
|
|
value = m.values.length > 0 ? m.values.split(/\s*,\s*/) : [];
|
|
}
|
|
|
|
result[match.name] = value;
|
|
}
|
|
|
|
return result;
|
|
};
|
|
|
|
/**
|
|
* Wraps each line of the string into <code/> tag with given style applied to it.
|
|
*
|
|
* @param {String} str Input string.
|
|
* @param {String} css Style name to apply to the string.
|
|
* @return {String} Returns input string with each line surrounded by <span/> tag.
|
|
*/
|
|
function wrapLinesWithCode(str, css)
|
|
{
|
|
if (str == null || str.length == 0 || str == '\n')
|
|
return str;
|
|
|
|
str = str.replace(/</g, '<');
|
|
|
|
// Replace two or more sequential spaces with leaving last space untouched.
|
|
str = str.replace(/ {2,}/g, function(m)
|
|
{
|
|
var spaces = '';
|
|
|
|
for (var i = 0; i < m.length - 1; i++)
|
|
spaces += sh.config.space;
|
|
|
|
return spaces + ' ';
|
|
});
|
|
|
|
// Split each line and apply <span class="...">...</span> to them so that
|
|
// leading spaces aren't included.
|
|
if (css != null)
|
|
str = eachLine(str, function(line)
|
|
{
|
|
if (line.length == 0)
|
|
return '';
|
|
|
|
var spaces = '';
|
|
|
|
line = line.replace(/^( | )+/, function(s)
|
|
{
|
|
spaces = s;
|
|
return '';
|
|
});
|
|
|
|
if (line.length == 0)
|
|
return spaces;
|
|
|
|
return spaces + '<code class="' + css + '">' + line + '</code>';
|
|
});
|
|
|
|
return str;
|
|
};
|
|
|
|
/**
|
|
* Pads number with zeros until it's length is the same as given length.
|
|
*
|
|
* @param {Number} number Number to pad.
|
|
* @param {Number} length Max string length with.
|
|
* @return {String} Returns a string padded with proper amount of '0'.
|
|
*/
|
|
function padNumber(number, length)
|
|
{
|
|
var result = number.toString();
|
|
|
|
while (result.length < length)
|
|
result = '0' + result;
|
|
|
|
return result;
|
|
};
|
|
|
|
/**
|
|
* Replaces tabs with spaces.
|
|
*
|
|
* @param {String} code Source code.
|
|
* @param {Number} tabSize Size of the tab.
|
|
* @return {String} Returns code with all tabs replaces by spaces.
|
|
*/
|
|
function processTabs(code, tabSize)
|
|
{
|
|
var tab = '';
|
|
|
|
for (var i = 0; i < tabSize; i++)
|
|
tab += ' ';
|
|
|
|
return code.replace(/\t/g, tab);
|
|
};
|
|
|
|
/**
|
|
* Replaces tabs with smart spaces.
|
|
*
|
|
* @param {String} code Code to fix the tabs in.
|
|
* @param {Number} tabSize Number of spaces in a column.
|
|
* @return {String} Returns code with all tabs replaces with roper amount of spaces.
|
|
*/
|
|
function processSmartTabs(code, tabSize)
|
|
{
|
|
var lines = splitLines(code),
|
|
tab = '\t',
|
|
spaces = ''
|
|
;
|
|
|
|
// Create a string with 1000 spaces to copy spaces from...
|
|
// It's assumed that there would be no indentation longer than that.
|
|
for (var i = 0; i < 50; i++)
|
|
spaces += ' '; // 20 spaces * 50
|
|
|
|
// This function inserts specified amount of spaces in the string
|
|
// where a tab is while removing that given tab.
|
|
function insertSpaces(line, pos, count)
|
|
{
|
|
return line.substr(0, pos)
|
|
+ spaces.substr(0, count)
|
|
+ line.substr(pos + 1, line.length) // pos + 1 will get rid of the tab
|
|
;
|
|
};
|
|
|
|
// Go through all the lines and do the 'smart tabs' magic.
|
|
code = eachLine(code, function(line)
|
|
{
|
|
if (line.indexOf(tab) == -1)
|
|
return line;
|
|
|
|
var pos = 0;
|
|
|
|
while ((pos = line.indexOf(tab)) != -1)
|
|
{
|
|
// This is pretty much all there is to the 'smart tabs' logic.
|
|
// Based on the position within the line and size of a tab,
|
|
// calculate the amount of spaces we need to insert.
|
|
var spaces = tabSize - pos % tabSize;
|
|
line = insertSpaces(line, pos, spaces);
|
|
}
|
|
|
|
return line;
|
|
});
|
|
|
|
return code;
|
|
};
|
|
|
|
/**
|
|
* Performs various string fixes based on configuration.
|
|
*/
|
|
function fixInputString(str)
|
|
{
|
|
var br = /<br\s*\/?>|<br\s*\/?>/gi;
|
|
|
|
if (sh.config.bloggerMode == true)
|
|
str = str.replace(br, '\n');
|
|
|
|
if (sh.config.stripBrs == true)
|
|
str = str.replace(br, '');
|
|
|
|
return str;
|
|
};
|
|
|
|
/**
|
|
* Removes all white space at the begining and end of a string.
|
|
*
|
|
* @param {String} str String to trim.
|
|
* @return {String} Returns string without leading and following white space characters.
|
|
*/
|
|
function trim(str)
|
|
{
|
|
return str.replace(/^\s+|\s+$/g, '');
|
|
};
|
|
|
|
/**
|
|
* Unindents a block of text by the lowest common indent amount.
|
|
* @param {String} str Text to unindent.
|
|
* @return {String} Returns unindented text block.
|
|
*/
|
|
function unindent(str)
|
|
{
|
|
var lines = splitLines(fixInputString(str)),
|
|
indents = new Array(),
|
|
regex = /^\s*/,
|
|
min = 1000
|
|
;
|
|
|
|
// go through every line and check for common number of indents
|
|
for (var i = 0; i < lines.length && min > 0; i++)
|
|
{
|
|
var line = lines[i];
|
|
|
|
if (trim(line).length == 0)
|
|
continue;
|
|
|
|
var matches = regex.exec(line);
|
|
|
|
// In the event that just one line doesn't have leading white space
|
|
// we can't unindent anything, so bail completely.
|
|
if (matches == null)
|
|
return str;
|
|
|
|
min = Math.min(matches[0].length, min);
|
|
}
|
|
|
|
// trim minimum common number of white space from the begining of every line
|
|
if (min > 0)
|
|
for (var i = 0; i < lines.length; i++)
|
|
lines[i] = lines[i].substr(min);
|
|
|
|
return lines.join('\n');
|
|
};
|
|
|
|
/**
|
|
* Callback method for Array.sort() which sorts matches by
|
|
* index position and then by length.
|
|
*
|
|
* @param {Match} m1 Left object.
|
|
* @param {Match} m2 Right object.
|
|
* @return {Number} Returns -1, 0 or -1 as a comparison result.
|
|
*/
|
|
function matchesSortCallback(m1, m2)
|
|
{
|
|
// sort matches by index first
|
|
if(m1.index < m2.index)
|
|
return -1;
|
|
else if(m1.index > m2.index)
|
|
return 1;
|
|
else
|
|
{
|
|
// if index is the same, sort by length
|
|
if(m1.length < m2.length)
|
|
return -1;
|
|
else if(m1.length > m2.length)
|
|
return 1;
|
|
}
|
|
|
|
return 0;
|
|
};
|
|
|
|
/**
|
|
* Executes given regular expression on provided code and returns all
|
|
* matches that are found.
|
|
*
|
|
* @param {String} code Code to execute regular expression on.
|
|
* @param {Object} regex Regular expression item info from <code>regexList</code> collection.
|
|
* @return {Array} Returns a list of Match objects.
|
|
*/
|
|
function getMatches(code, regexInfo)
|
|
{
|
|
function defaultAdd(match, regexInfo)
|
|
{
|
|
return match[0];
|
|
};
|
|
|
|
var index = 0,
|
|
match = null,
|
|
matches = [],
|
|
func = regexInfo.func ? regexInfo.func : defaultAdd
|
|
;
|
|
|
|
while((match = regexInfo.regex.exec(code)) != null)
|
|
{
|
|
var resultMatch = func(match, regexInfo);
|
|
|
|
if (typeof(resultMatch) == 'string')
|
|
resultMatch = [new sh.Match(resultMatch, match.index, regexInfo.css)];
|
|
|
|
matches = matches.concat(resultMatch);
|
|
}
|
|
|
|
return matches;
|
|
};
|
|
|
|
/**
|
|
* Turns all URLs in the code into <a/> tags.
|
|
* @param {String} code Input code.
|
|
* @return {String} Returns code with </a> tags.
|
|
*/
|
|
function processUrls(code)
|
|
{
|
|
var gt = /(.*)((>|<).*)/;
|
|
|
|
return code.replace(sh.regexLib.url, function(m)
|
|
{
|
|
var suffix = '',
|
|
match = null
|
|
;
|
|
|
|
// We include < and > in the URL for the common cases like <http://google.com>
|
|
// The problem is that they get transformed into <http://google.com>
|
|
// Where as > easily looks like part of the URL string.
|
|
|
|
if (match = gt.exec(m))
|
|
{
|
|
m = match[1];
|
|
suffix = match[2];
|
|
}
|
|
|
|
return '<a href="' + m + '">' + m + '</a>' + suffix;
|
|
});
|
|
};
|
|
|
|
/**
|
|
* Finds all <SCRIPT TYPE="syntaxhighlighter" /> elementss.
|
|
* @return {Array} Returns array of all found SyntaxHighlighter tags.
|
|
*/
|
|
function getSyntaxHighlighterScriptTags()
|
|
{
|
|
var tags = document.getElementsByTagName('script'),
|
|
result = []
|
|
;
|
|
|
|
for (var i = 0; i < tags.length; i++)
|
|
if (tags[i].type == 'syntaxhighlighter')
|
|
result.push(tags[i]);
|
|
|
|
return result;
|
|
};
|
|
|
|
/**
|
|
* Strips <![CDATA[]]> from <SCRIPT /> content because it should be used
|
|
* there in most cases for XHTML compliance.
|
|
* @param {String} original Input code.
|
|
* @return {String} Returns code without leading <![CDATA[]]> tags.
|
|
*/
|
|
function stripCData(original)
|
|
{
|
|
var left = '<![CDATA[',
|
|
right = ']]>',
|
|
// for some reason IE inserts some leading blanks here
|
|
copy = trim(original),
|
|
changed = false,
|
|
leftLength = left.length,
|
|
rightLength = right.length
|
|
;
|
|
|
|
if (copy.indexOf(left) == 0)
|
|
{
|
|
copy = copy.substring(leftLength);
|
|
changed = true;
|
|
}
|
|
|
|
var copyLength = copy.length;
|
|
|
|
if (copy.indexOf(right) == copyLength - rightLength)
|
|
{
|
|
copy = copy.substring(0, copyLength - rightLength);
|
|
changed = true;
|
|
}
|
|
|
|
return changed ? copy : original;
|
|
};
|
|
|
|
|
|
/**
|
|
* Quick code mouse double click handler.
|
|
*/
|
|
function quickCodeHandler(e)
|
|
{
|
|
var target = e.target,
|
|
highlighterDiv = findParentElement(target, '.syntaxhighlighter'),
|
|
container = findParentElement(target, '.container'),
|
|
textarea = document.createElement('textarea'),
|
|
highlighter
|
|
;
|
|
|
|
if (!container || !highlighterDiv || findElement(container, 'textarea'))
|
|
return;
|
|
|
|
highlighter = getHighlighterById(highlighterDiv.id);
|
|
|
|
// add source class name
|
|
addClass(highlighterDiv, 'source');
|
|
|
|
// Have to go over each line and grab it's text, can't just do it on the
|
|
// container because Firefox loses all \n where as Webkit doesn't.
|
|
var lines = container.childNodes,
|
|
code = []
|
|
;
|
|
|
|
for (var i = 0; i < lines.length; i++)
|
|
code.push(lines[i].innerText || lines[i].textContent);
|
|
|
|
// using \r instead of \r or \r\n makes this work equally well on IE, FF and Webkit
|
|
code = code.join('\r');
|
|
|
|
// For Webkit browsers, replace nbsp with a breaking space
|
|
code = code.replace(/\u00a0/g, " ");
|
|
|
|
// inject <textarea/> tag
|
|
textarea.appendChild(document.createTextNode(code));
|
|
container.appendChild(textarea);
|
|
|
|
// preselect all text
|
|
textarea.focus();
|
|
textarea.select();
|
|
|
|
// set up handler for lost focus
|
|
attachEvent(textarea, 'blur', function(e)
|
|
{
|
|
textarea.parentNode.removeChild(textarea);
|
|
removeClass(highlighterDiv, 'source');
|
|
});
|
|
};
|
|
|
|
/**
|
|
* Match object.
|
|
*/
|
|
sh.Match = function(value, index, css)
|
|
{
|
|
this.value = value;
|
|
this.index = index;
|
|
this.length = value.length;
|
|
this.css = css;
|
|
this.brushName = null;
|
|
};
|
|
|
|
sh.Match.prototype.toString = function()
|
|
{
|
|
return this.value;
|
|
};
|
|
|
|
/**
|
|
* Simulates HTML code with a scripting language embedded.
|
|
*
|
|
* @param {String} scriptBrushName Brush name of the scripting language.
|
|
*/
|
|
sh.HtmlScript = function(scriptBrushName)
|
|
{
|
|
var brushClass = findBrush(scriptBrushName),
|
|
scriptBrush,
|
|
xmlBrush = new sh.brushes.Xml(),
|
|
bracketsRegex = null,
|
|
ref = this,
|
|
methodsToExpose = 'getDiv getHtml init'.split(' ')
|
|
;
|
|
|
|
if (brushClass == null)
|
|
return;
|
|
|
|
scriptBrush = new brushClass();
|
|
|
|
for(var i = 0; i < methodsToExpose.length; i++)
|
|
// make a closure so we don't lose the name after i changes
|
|
(function() {
|
|
var name = methodsToExpose[i];
|
|
|
|
ref[name] = function()
|
|
{
|
|
return xmlBrush[name].apply(xmlBrush, arguments);
|
|
};
|
|
})();
|
|
|
|
if (scriptBrush.htmlScript == null)
|
|
{
|
|
alert(sh.config.strings.brushNotHtmlScript + scriptBrushName);
|
|
return;
|
|
}
|
|
|
|
xmlBrush.regexList.push(
|
|
{ regex: scriptBrush.htmlScript.code, func: process }
|
|
);
|
|
|
|
function offsetMatches(matches, offset)
|
|
{
|
|
for (var j = 0; j < matches.length; j++)
|
|
matches[j].index += offset;
|
|
}
|
|
|
|
function process(match, info)
|
|
{
|
|
var code = match.code,
|
|
matches = [],
|
|
regexList = scriptBrush.regexList,
|
|
offset = match.index + match.left.length,
|
|
htmlScript = scriptBrush.htmlScript,
|
|
result
|
|
;
|
|
|
|
// add all matches from the code
|
|
for (var i = 0; i < regexList.length; i++)
|
|
{
|
|
result = getMatches(code, regexList[i]);
|
|
offsetMatches(result, offset);
|
|
matches = matches.concat(result);
|
|
}
|
|
|
|
// add left script bracket
|
|
if (htmlScript.left != null && match.left != null)
|
|
{
|
|
result = getMatches(match.left, htmlScript.left);
|
|
offsetMatches(result, match.index);
|
|
matches = matches.concat(result);
|
|
}
|
|
|
|
// add right script bracket
|
|
if (htmlScript.right != null && match.right != null)
|
|
{
|
|
result = getMatches(match.right, htmlScript.right);
|
|
offsetMatches(result, match.index + match[0].lastIndexOf(match.right));
|
|
matches = matches.concat(result);
|
|
}
|
|
|
|
for (var j = 0; j < matches.length; j++)
|
|
matches[j].brushName = brushClass.brushName;
|
|
|
|
return matches;
|
|
}
|
|
};
|
|
|
|
/**
|
|
* Main Highlither class.
|
|
* @constructor
|
|
*/
|
|
sh.Highlighter = function()
|
|
{
|
|
// not putting any code in here because of the prototype inheritance
|
|
};
|
|
|
|
sh.Highlighter.prototype = {
|
|
/**
|
|
* Returns value of the parameter passed to the highlighter.
|
|
* @param {String} name Name of the parameter.
|
|
* @param {Object} defaultValue Default value.
|
|
* @return {Object} Returns found value or default value otherwise.
|
|
*/
|
|
getParam: function(name, defaultValue)
|
|
{
|
|
var result = this.params[name];
|
|
return toBoolean(result == null ? defaultValue : result);
|
|
},
|
|
|
|
/**
|
|
* Shortcut to document.createElement().
|
|
* @param {String} name Name of the element to create (DIV, A, etc).
|
|
* @return {HTMLElement} Returns new HTML element.
|
|
*/
|
|
create: function(name)
|
|
{
|
|
return document.createElement(name);
|
|
},
|
|
|
|
/**
|
|
* Applies all regular expression to the code and stores all found
|
|
* matches in the `this.matches` array.
|
|
* @param {Array} regexList List of regular expressions.
|
|
* @param {String} code Source code.
|
|
* @return {Array} Returns list of matches.
|
|
*/
|
|
findMatches: function(regexList, code)
|
|
{
|
|
var result = [];
|
|
|
|
if (regexList != null)
|
|
for (var i = 0; i < regexList.length; i++)
|
|
// BUG: length returns len+1 for array if methods added to prototype chain (oising@gmail.com)
|
|
if (typeof (regexList[i]) == "object")
|
|
result = result.concat(getMatches(code, regexList[i]));
|
|
|
|
// sort and remove nested the matches
|
|
return this.removeNestedMatches(result.sort(matchesSortCallback));
|
|
},
|
|
|
|
/**
|
|
* Checks to see if any of the matches are inside of other matches.
|
|
* This process would get rid of highligted strings inside comments,
|
|
* keywords inside strings and so on.
|
|
*/
|
|
removeNestedMatches: function(matches)
|
|
{
|
|
// Optimized by Jose Prado (http://joseprado.com)
|
|
for (var i = 0; i < matches.length; i++)
|
|
{
|
|
if (matches[i] === null)
|
|
continue;
|
|
|
|
var itemI = matches[i],
|
|
itemIEndPos = itemI.index + itemI.length
|
|
;
|
|
|
|
for (var j = i + 1; j < matches.length && matches[i] !== null; j++)
|
|
{
|
|
var itemJ = matches[j];
|
|
|
|
if (itemJ === null)
|
|
continue;
|
|
else if (itemJ.index > itemIEndPos)
|
|
break;
|
|
else if (itemJ.index == itemI.index && itemJ.length > itemI.length)
|
|
matches[i] = null;
|
|
else if (itemJ.index >= itemI.index && itemJ.index < itemIEndPos)
|
|
matches[j] = null;
|
|
}
|
|
}
|
|
|
|
return matches;
|
|
},
|
|
|
|
/**
|
|
* Creates an array containing integer line numbers starting from the 'first-line' param.
|
|
* @return {Array} Returns array of integers.
|
|
*/
|
|
figureOutLineNumbers: function(code)
|
|
{
|
|
var lines = [],
|
|
firstLine = parseInt(this.getParam('first-line'))
|
|
;
|
|
|
|
eachLine(code, function(line, index)
|
|
{
|
|
lines.push(index + firstLine);
|
|
});
|
|
|
|
return lines;
|
|
},
|
|
|
|
/**
|
|
* Determines if specified line number is in the highlighted list.
|
|
*/
|
|
isLineHighlighted: function(lineNumber)
|
|
{
|
|
var list = this.getParam('highlight', []);
|
|
|
|
if (typeof(list) != 'object' && list.push == null)
|
|
list = [ list ];
|
|
|
|
return indexOf(list, lineNumber.toString()) != -1;
|
|
},
|
|
|
|
/**
|
|
* Generates HTML markup for a single line of code while determining alternating line style.
|
|
* @param {Integer} lineNumber Line number.
|
|
* @param {String} code Line HTML markup.
|
|
* @return {String} Returns HTML markup.
|
|
*/
|
|
getLineHtml: function(lineIndex, lineNumber, code)
|
|
{
|
|
var classes = [
|
|
'line',
|
|
'number' + lineNumber,
|
|
'index' + lineIndex,
|
|
'alt' + (lineNumber % 2 == 0 ? 1 : 2).toString()
|
|
];
|
|
|
|
if (this.isLineHighlighted(lineNumber))
|
|
classes.push('highlighted');
|
|
|
|
if (lineNumber == 0)
|
|
classes.push('break');
|
|
|
|
return '<div class="' + classes.join(' ') + '">' + code + '</div>';
|
|
},
|
|
|
|
/**
|
|
* Generates HTML markup for line number column.
|
|
* @param {String} code Complete code HTML markup.
|
|
* @param {Array} lineNumbers Calculated line numbers.
|
|
* @return {String} Returns HTML markup.
|
|
*/
|
|
getLineNumbersHtml: function(code, lineNumbers)
|
|
{
|
|
var html = '',
|
|
count = splitLines(code).length,
|
|
firstLine = parseInt(this.getParam('first-line')),
|
|
pad = this.getParam('pad-line-numbers')
|
|
;
|
|
|
|
if (pad == true)
|
|
pad = (firstLine + count - 1).toString().length;
|
|
else if (isNaN(pad) == true)
|
|
pad = 0;
|
|
|
|
for (var i = 0; i < count; i++)
|
|
{
|
|
var lineNumber = lineNumbers ? lineNumbers[i] : firstLine + i,
|
|
code = lineNumber == 0 ? sh.config.space : padNumber(lineNumber, pad)
|
|
;
|
|
|
|
html += this.getLineHtml(i, lineNumber, code);
|
|
}
|
|
|
|
return html;
|
|
},
|
|
|
|
/**
|
|
* Splits block of text into individual DIV lines.
|
|
* @param {String} code Code to highlight.
|
|
* @param {Array} lineNumbers Calculated line numbers.
|
|
* @return {String} Returns highlighted code in HTML form.
|
|
*/
|
|
getCodeLinesHtml: function(html, lineNumbers)
|
|
{
|
|
html = trim(html);
|
|
|
|
var lines = splitLines(html),
|
|
padLength = this.getParam('pad-line-numbers'),
|
|
firstLine = parseInt(this.getParam('first-line')),
|
|
html = '',
|
|
brushName = this.getParam('brush')
|
|
;
|
|
|
|
for (var i = 0; i < lines.length; i++)
|
|
{
|
|
var line = lines[i],
|
|
indent = /^( |\s)+/.exec(line),
|
|
spaces = null,
|
|
lineNumber = lineNumbers ? lineNumbers[i] : firstLine + i;
|
|
;
|
|
|
|
if (indent != null)
|
|
{
|
|
spaces = indent[0].toString();
|
|
line = line.substr(spaces.length);
|
|
spaces = spaces.replace(' ', sh.config.space);
|
|
}
|
|
|
|
line = trim(line);
|
|
|
|
if (line.length == 0)
|
|
line = sh.config.space;
|
|
|
|
html += this.getLineHtml(
|
|
i,
|
|
lineNumber,
|
|
(spaces != null ? '<code class="' + brushName + ' spaces">' + spaces + '</code>' : '') + line
|
|
);
|
|
}
|
|
|
|
return html;
|
|
},
|
|
|
|
/**
|
|
* Returns HTML for the table title or empty string if title is null.
|
|
*/
|
|
getTitleHtml: function(title)
|
|
{
|
|
return title ? '<caption>' + title + '</caption>' : '';
|
|
},
|
|
|
|
/**
|
|
* Finds all matches in the source code.
|
|
* @param {String} code Source code to process matches in.
|
|
* @param {Array} matches Discovered regex matches.
|
|
* @return {String} Returns formatted HTML with processed mathes.
|
|
*/
|
|
getMatchesHtml: function(code, matches)
|
|
{
|
|
var pos = 0,
|
|
result = '',
|
|
brushName = this.getParam('brush', '')
|
|
;
|
|
|
|
function getBrushNameCss(match)
|
|
{
|
|
var result = match ? (match.brushName || brushName) : brushName;
|
|
return result ? result + ' ' : '';
|
|
};
|
|
|
|
// Finally, go through the final list of matches and pull the all
|
|
// together adding everything in between that isn't a match.
|
|
for (var i = 0; i < matches.length; i++)
|
|
{
|
|
var match = matches[i],
|
|
matchBrushName
|
|
;
|
|
|
|
if (match === null || match.length === 0)
|
|
continue;
|
|
|
|
matchBrushName = getBrushNameCss(match);
|
|
|
|
result += wrapLinesWithCode(code.substr(pos, match.index - pos), matchBrushName + 'plain')
|
|
+ wrapLinesWithCode(match.value, matchBrushName + match.css)
|
|
;
|
|
|
|
pos = match.index + match.length + (match.offset || 0);
|
|
}
|
|
|
|
// don't forget to add whatever's remaining in the string
|
|
result += wrapLinesWithCode(code.substr(pos), getBrushNameCss() + 'plain');
|
|
|
|
return result;
|
|
},
|
|
|
|
/**
|
|
* Generates HTML markup for the whole syntax highlighter.
|
|
* @param {String} code Source code.
|
|
* @return {String} Returns HTML markup.
|
|
*/
|
|
getHtml: function(code)
|
|
{
|
|
var html = '',
|
|
classes = [ 'syntaxhighlighter' ],
|
|
tabSize,
|
|
matches,
|
|
lineNumbers
|
|
;
|
|
|
|
// process light mode
|
|
if (this.getParam('light') == true)
|
|
this.params.toolbar = this.params.gutter = false;
|
|
|
|
className = 'syntaxhighlighter';
|
|
|
|
if (this.getParam('collapse') == true)
|
|
classes.push('collapsed');
|
|
|
|
if ((gutter = this.getParam('gutter')) == false)
|
|
classes.push('nogutter');
|
|
|
|
// add custom user style name
|
|
classes.push(this.getParam('class-name'));
|
|
|
|
// add brush alias to the class name for custom CSS
|
|
classes.push(this.getParam('brush'));
|
|
|
|
code = trimFirstAndLastLines(code)
|
|
.replace(/\r/g, ' ') // IE lets these buggers through
|
|
;
|
|
|
|
tabSize = this.getParam('tab-size');
|
|
|
|
// replace tabs with spaces
|
|
code = this.getParam('smart-tabs') == true
|
|
? processSmartTabs(code, tabSize)
|
|
: processTabs(code, tabSize)
|
|
;
|
|
|
|
// unindent code by the common indentation
|
|
if (this.getParam('unindent'))
|
|
code = unindent(code);
|
|
|
|
if (gutter)
|
|
lineNumbers = this.figureOutLineNumbers(code);
|
|
|
|
// find matches in the code using brushes regex list
|
|
matches = this.findMatches(this.regexList, code);
|
|
// processes found matches into the html
|
|
html = this.getMatchesHtml(code, matches);
|
|
// finally, split all lines so that they wrap well
|
|
html = this.getCodeLinesHtml(html, lineNumbers);
|
|
|
|
// finally, process the links
|
|
if (this.getParam('auto-links'))
|
|
html = processUrls(html);
|
|
|
|
if (typeof(navigator) != 'undefined' && navigator.userAgent && navigator.userAgent.match(/MSIE/))
|
|
classes.push('ie');
|
|
|
|
html =
|
|
'<div id="' + getHighlighterId(this.id) + '" class="' + classes.join(' ') + '">'
|
|
+ (this.getParam('toolbar') ? sh.toolbar.getHtml(this) : '')
|
|
+ '<table border="0" cellpadding="0" cellspacing="0">'
|
|
+ this.getTitleHtml(this.getParam('title'))
|
|
+ '<tbody>'
|
|
+ '<tr>'
|
|
+ (gutter ? '<td class="gutter">' + this.getLineNumbersHtml(code) + '</td>' : '')
|
|
+ '<td class="code">'
|
|
+ '<div class="container">'
|
|
+ html
|
|
+ '</div>'
|
|
+ '</td>'
|
|
+ '</tr>'
|
|
+ '</tbody>'
|
|
+ '</table>'
|
|
+ '</div>'
|
|
;
|
|
|
|
return html;
|
|
},
|
|
|
|
/**
|
|
* Highlights the code and returns complete HTML.
|
|
* @param {String} code Code to highlight.
|
|
* @return {Element} Returns container DIV element with all markup.
|
|
*/
|
|
getDiv: function(code)
|
|
{
|
|
if (code === null)
|
|
code = '';
|
|
|
|
this.code = code;
|
|
|
|
var div = this.create('div');
|
|
|
|
// create main HTML
|
|
div.innerHTML = this.getHtml(code);
|
|
|
|
// set up click handlers
|
|
if (this.getParam('toolbar'))
|
|
attachEvent(findElement(div, '.toolbar'), 'click', sh.toolbar.handler);
|
|
|
|
if (this.getParam('quick-code'))
|
|
attachEvent(findElement(div, '.code'), 'dblclick', quickCodeHandler);
|
|
|
|
return div;
|
|
},
|
|
|
|
/**
|
|
* Initializes the highlighter/brush.
|
|
*
|
|
* Constructor isn't used for initialization so that nothing executes during necessary
|
|
* `new SyntaxHighlighter.Highlighter()` call when setting up brush inheritence.
|
|
*
|
|
* @param {Hash} params Highlighter parameters.
|
|
*/
|
|
init: function(params)
|
|
{
|
|
this.id = guid();
|
|
|
|
// register this instance in the highlighters list
|
|
storeHighlighter(this);
|
|
|
|
// local params take precedence over defaults
|
|
this.params = merge(sh.defaults, params || {})
|
|
|
|
// process light mode
|
|
if (this.getParam('light') == true)
|
|
this.params.toolbar = this.params.gutter = false;
|
|
},
|
|
|
|
/**
|
|
* Converts space separated list of keywords into a regular expression string.
|
|
* @param {String} str Space separated keywords.
|
|
* @return {String} Returns regular expression string.
|
|
*/
|
|
getKeywords: function(str)
|
|
{
|
|
str = str
|
|
.replace(/^\s+|\s+$/g, '')
|
|
.replace(/\s+/g, '|')
|
|
;
|
|
|
|
return '\\b(?:' + str + ')\\b';
|
|
},
|
|
|
|
/**
|
|
* Makes a brush compatible with the `html-script` functionality.
|
|
* @param {Object} regexGroup Object containing `left` and `right` regular expressions.
|
|
*/
|
|
forHtmlScript: function(regexGroup)
|
|
{
|
|
var regex = { 'end' : regexGroup.right.source };
|
|
|
|
if(regexGroup.eof)
|
|
regex.end = "(?:(?:" + regex.end + ")|$)";
|
|
|
|
this.htmlScript = {
|
|
left : { regex: regexGroup.left, css: 'script' },
|
|
right : { regex: regexGroup.right, css: 'script' },
|
|
code : new XRegExp(
|
|
"(?<left>" + regexGroup.left.source + ")" +
|
|
"(?<code>.*?)" +
|
|
"(?<right>" + regex.end + ")",
|
|
"sgi"
|
|
)
|
|
};
|
|
}
|
|
}; // end of Highlighter
|
|
|
|
return sh;
|
|
}(); // end of anonymous function
|
|
|
|
// CommonJS
|
|
typeof(exports) != 'undefined' ? exports.SyntaxHighlighter = SyntaxHighlighter : null;/**
|
|
* SyntaxHighlighter
|
|
* http://alexgorbatchev.com/SyntaxHighlighter
|
|
*
|
|
* SyntaxHighlighter is donationware. If you are using it, please donate.
|
|
* http://alexgorbatchev.com/SyntaxHighlighter/donate.html
|
|
*
|
|
* @version
|
|
* 3.0.83 (July 02 2010)
|
|
*
|
|
* @copyright
|
|
* Copyright (C) 2004-2010 Alex Gorbatchev.
|
|
*
|
|
* @license
|
|
* Dual licensed under the MIT and GPL licenses.
|
|
*/
|
|
;(function()
|
|
{
|
|
// CommonJS
|
|
typeof(require) != 'undefined' ? SyntaxHighlighter = require('shCore').SyntaxHighlighter : null;
|
|
|
|
function Brush()
|
|
{
|
|
var keywords = 'break case catch continue ' +
|
|
'default delete do else false ' +
|
|
'for function if in instanceof ' +
|
|
'new null return super switch ' +
|
|
'this throw true try typeof var while with'
|
|
;
|
|
|
|
var datatablesAPI =
|
|
'fnSettings fnDraw fnFilter fnSort fnAddData fnDeleteRow fnClearTable fnGetData '+
|
|
'fnGetPosition fnGetNodes fnOpen fnClose fnUpdate fnSetColumnVis fnVersionCheck '+
|
|
'fnPageChange fnSortListener fnDestroy fnAdjustColumnSizing';
|
|
|
|
var r = SyntaxHighlighter.regexLib;
|
|
|
|
this.regexList = [
|
|
{ regex: r.multiLineDoubleQuotedString, css: 'string' }, // double quoted strings
|
|
{ regex: r.multiLineSingleQuotedString, css: 'string' }, // single quoted strings
|
|
{ regex: r.singleLineCComments, css: 'comments' }, // one line comments
|
|
{ regex: r.multiLineCComments, css: 'comments' }, // multiline comments
|
|
{ regex: /\s*#.*/gm, css: 'preprocessor' }, // preprocessor tags like #region and #endregion
|
|
{ regex: new RegExp(this.getKeywords(keywords), 'gm'), css: 'keyword' }, // keywords
|
|
{ regex: new RegExp(this.getKeywords(datatablesAPI), 'gm'), css: 'dtapi' } // DataTables API methods
|
|
];
|
|
|
|
this.forHtmlScript(r.scriptScriptTags);
|
|
};
|
|
|
|
Brush.prototype = new SyntaxHighlighter.Highlighter();
|
|
Brush.aliases = ['js', 'jscript', 'javascript'];
|
|
|
|
SyntaxHighlighter.brushes.JScript = Brush;
|
|
|
|
// CommonJS
|
|
typeof(exports) != 'undefined' ? exports.Brush = Brush : null;
|
|
})();
|
|
|
|
|
|
|
|
|
|
|
|
/**
|
|
* SyntaxHighlighter
|
|
* http://alexgorbatchev.com/SyntaxHighlighter
|
|
*
|
|
* SyntaxHighlighter is donationware. If you are using it, please donate.
|
|
* http://alexgorbatchev.com/SyntaxHighlighter/donate.html
|
|
*
|
|
* @version
|
|
* 3.0.83 (July 02 2010)
|
|
*
|
|
* @copyright
|
|
* Copyright (C) 2004-2010 Alex Gorbatchev.
|
|
*
|
|
* @license
|
|
* Dual licensed under the MIT and GPL licenses.
|
|
*/
|
|
;(function()
|
|
{
|
|
// CommonJS
|
|
typeof(require) != 'undefined' ? SyntaxHighlighter = require('shCore').SyntaxHighlighter : null;
|
|
|
|
function Brush()
|
|
{
|
|
function process(match, regexInfo)
|
|
{
|
|
var constructor = SyntaxHighlighter.Match,
|
|
code = match[0],
|
|
tag = new XRegExp('(<|<)[\\s\\/\\?]*(?<name>[:\\w-\\.]+)', 'xg').exec(code),
|
|
result = []
|
|
;
|
|
|
|
if (match.attributes != null)
|
|
{
|
|
var attributes,
|
|
regex = new XRegExp('(?<name> [\\w:\\-\\.]+)' +
|
|
'\\s*=\\s*' +
|
|
'(?<value> ".*?"|\'.*?\'|\\w+)',
|
|
'xg');
|
|
|
|
while ((attributes = regex.exec(code)) != null)
|
|
{
|
|
result.push(new constructor(attributes.name, match.index + attributes.index, 'color1'));
|
|
result.push(new constructor(attributes.value, match.index + attributes.index + attributes[0].indexOf(attributes.value), 'string'));
|
|
}
|
|
}
|
|
|
|
if (tag != null)
|
|
result.push(
|
|
new constructor(tag.name, match.index + tag[0].indexOf(tag.name), 'keyword')
|
|
);
|
|
|
|
return result;
|
|
}
|
|
|
|
this.regexList = [
|
|
{ regex: new XRegExp('(\\<|<)\\!\\[[\\w\\s]*?\\[(.|\\s)*?\\]\\](\\>|>)', 'gm'), css: 'color2' }, // <![ ... [ ... ]]>
|
|
{ regex: SyntaxHighlighter.regexLib.xmlComments, css: 'comments' }, // <!-- ... -->
|
|
{ regex: new XRegExp('(<|<)[\\s\\/\\?]*(\\w+)(?<attributes>.*?)[\\s\\/\\?]*(>|>)', 'sg'), func: process }
|
|
];
|
|
};
|
|
|
|
Brush.prototype = new SyntaxHighlighter.Highlighter();
|
|
Brush.aliases = ['xml', 'xhtml', 'xslt', 'html'];
|
|
|
|
SyntaxHighlighter.brushes.Xml = Brush;
|
|
|
|
// CommonJS
|
|
typeof(exports) != 'undefined' ? exports.Brush = Brush : null;
|
|
})();
|
|
|
|
|
|
|
|
/* Self initialise */
|
|
SyntaxHighlighter.all();
|