欢迎您访问程序员文章站本站旨在为大家提供分享程序员计算机编程知识!
您现在的位置是: 首页

editplus json文本格式化

程序员文章站 2022-03-07 17:20:00
...

json格式化脚本下载:https://download.csdn.net/download/sanzhongguren/10367602

Tools->Configure User Tools->User tools->Add Tool->Program

editplus json文本格式化

Menu text:名称随意

Command:Cscript.exe /nologo C:\program_files\EditPlus\jsFormatter.js

如果提示找不到文件可以尝试用英文引号将路径引起来

editplus json文本格式化

格式化之前

editplus json文本格式化

CTRL+1 格式化之后

editplus json文本格式化

没有下载积分的可以复制以下脚本代码

//初始化选项
var indent_size = 1;//缩进空格数, 为1时使用制表符缩进
var indent_char = ' ';//缩进字符
var preserve_newlines = false;//是否保留空行, 默认不保留

//程序开始
var input = "";
while(!WScript.StdIn.AtEndOfStream)
{
   input += WScript.StdIn.ReadAll();
}

if (indent_size == 1) {
   indent_char = '\t';
}

var js_source = input.replace(/^\s+/, '');
var formated_code='';
if (js_source && js_source[0] !== '<') {
        formated_code =js_beautify(js_source, {indent_size: indent_size, indent_char: indent_char, preserve_newlines:preserve_newlines});
    }
if(!formated_code.length==0)
	WScript.Echo(formated_code);
else
	WScript.Echo('Are you sure your input is javascript source file?');

/*

 JS Beautifier
---------------
  $Date$
  $Revision$


  Written by Einars Lielmanis, <aaa@qq.com>
      http://elfz.laacz.lv/beautify/

  Originally converted to javascript by Vital, <aaa@qq.com>
      http://my.opera.com/Vital/blog/2007/11/21/javascript-beautify-on-javascript-translated


  You are free to use this in any way you want, in case you find this useful or working for you.

  Usage:
    js_beautify(js_source_text);
    js_beautify(js_source_text, options);

  The options are:
    indent_size (default 4) — indentation size,
    indent_char (default space) — character to indent with,
    preserve_newlines (default true) — whether existing line breaks should be preserved,
    indent_level (default 0)  — initial indentation level, you probably won't need this ever,

    e.g

    js_beautify(js_source_text, {indent_size: 1, indent_char: '\t'});


*/



function js_beautify(js_source_text, options)
{

    var input, output, token_text, last_type, last_text, last_word, current_mode, modes, indent_string;
    var whitespace, wordchar, punct, parser_pos, line_starters, in_case;
    var prefix, token_type, do_block_just_closed, var_line, var_line_tainted, if_line_flag;
    var indent_level;


    var options               = options || {};
    var opt_indent_size       = options['indent_size'] || 4;
    var opt_indent_char       = options['indent_char'] || ' ';
    var opt_preserve_newlines =
        typeof options['preserve_newlines'] === 'undefined' ? true : options['preserve_newlines'];
    var opt_indent_level      = options['indent_level'] || 0; // starting indentation


    function trim_output()
    {
        while (output.length && (output[output.length - 1] === ' ' || output[output.length - 1] === indent_string)) {
            output.pop();
        }
    }

    function print_newline(ignore_repeated)
    {

        ignore_repeated = typeof ignore_repeated === 'undefined' ? true: ignore_repeated;

        if_line_flag = false;
        trim_output();

        if (!output.length) {
            return; // no newline on start of file
        }

        if (output[output.length - 1] !== "\n" || !ignore_repeated) {
            output.push("\n");
        }
        for (var i = 0; i < indent_level; i++) {
            output.push(indent_string);
        }
    }



    function print_space()
    {
        var last_output = output.length ? output[output.length - 1] : ' ';
        if (last_output !== ' ' && last_output !== '\n' && last_output !== indent_string) { // prevent occassional duplicate space
            output.push(' ');
        }
    }


    function print_token()
    {
        output.push(token_text);
    }

    function indent()
    {
        indent_level++;
    }


    function unindent()
    {
        if (indent_level) {
            indent_level--;
        }
    }


    function remove_indent()
    {
        if (output.length && output[output.length - 1] === indent_string) {
            output.pop();
        }
    }


    function set_mode(mode)
    {
        modes.push(current_mode);
        current_mode = mode;
    }


    function restore_mode()
    {
        do_block_just_closed = current_mode === 'DO_BLOCK';
        current_mode = modes.pop();
    }


    function in_array(what, arr)
    {
        for (var i = 0; i < arr.length; i++)
        {
            if (arr[i] === what) {
                return true;
            }
        }
        return false;
    }



    function get_next_token()
    {
        var n_newlines = 0;
        var c = '';

        do {
            if (parser_pos >= input.length) {
                return ['', 'TK_EOF'];
            }
            c = input.charAt(parser_pos);

            parser_pos += 1;
            if (c === "\n") {
                n_newlines += 1;
            }
        }
        while (in_array(c, whitespace));

        var wanted_newline = false;

        if (opt_preserve_newlines) {
            if (n_newlines > 1) {
                for (var i = 0; i < 2; i++) {
                    print_newline(i === 0);
                }
            }
            wanted_newline = (n_newlines === 1);
        }


        if (in_array(c, wordchar)) {
            if (parser_pos < input.length) {
                while (in_array(input.charAt(parser_pos), wordchar)) {
                    c += input.charAt(parser_pos);
                    parser_pos += 1;
                    if (parser_pos === input.length) {
                        break;
                    }
                }
            }

            // small and surprisingly unugly hack for 1E-10 representation
            if (parser_pos !== input.length && c.match(/^[0-9]+[Ee]$/) && input.charAt(parser_pos) === '-') {
                parser_pos += 1;

                var t = get_next_token(parser_pos);
                c += '-' + t[0];
                return [c, 'TK_WORD'];
            }

            if (c === 'in') { // hack for 'in' operator
                return [c, 'TK_OPERATOR'];
            }
            if (wanted_newline && last_type !== 'TK_OPERATOR' && !if_line_flag) {
                print_newline();
            }
            return [c, 'TK_WORD'];
        }

        if (c === '(' || c === '[') {
            return [c, 'TK_START_EXPR'];
        }

        if (c === ')' || c === ']') {
            return [c, 'TK_END_EXPR'];
        }

        if (c === '{') {
            return [c, 'TK_START_BLOCK'];
        }

        if (c === '}') {
            return [c, 'TK_END_BLOCK'];
        }

        if (c === ';') {
            return [c, 'TK_SEMICOLON'];
        }

        if (c === '/') {
            var comment = '';
            // peek for comment /* ... */
            if (input.charAt(parser_pos) === '*') {
                parser_pos += 1;
                if (parser_pos < input.length) {
                    while (! (input.charAt(parser_pos) === '*' && input.charAt(parser_pos + 1) && input.charAt(parser_pos + 1) === '/') && parser_pos < input.length) {
                        comment += input.charAt(parser_pos);
                        parser_pos += 1;
                        if (parser_pos >= input.length) {
                            break;
                        }
                    }
                }
                parser_pos += 2;
                return ['/*' + comment + '*/', 'TK_BLOCK_COMMENT'];
            }
            // peek for comment // ...
            if (input.charAt(parser_pos) === '/') {
                comment = c;
                while (input.charAt(parser_pos) !== "\x0d" && input.charAt(parser_pos) !== "\x0a") {
                    comment += input.charAt(parser_pos);
                    parser_pos += 1;
                    if (parser_pos >= input.length) {
                        break;
                    }
                }
                parser_pos += 1;
                if (wanted_newline) {
                    print_newline();
                }
                return [comment, 'TK_COMMENT'];
            }

        }

        if (c === "'" || // string
        c === '"' || // string
        (c === '/' &&
        ((last_type === 'TK_WORD' && last_text === 'return') || (last_type === 'TK_START_EXPR' || last_type === 'TK_END_BLOCK' || last_type === 'TK_OPERATOR' || last_type === 'TK_EOF' || last_type === 'TK_SEMICOLON')))) { // regexp
            var sep = c;
            var esc = false;
            var resulting_string = '';

            if (parser_pos < input.length) {

                while (esc || input.charAt(parser_pos) !== sep) {
                    resulting_string += input.charAt(parser_pos);
                    if (!esc) {
                        esc = input.charAt(parser_pos) === '\\';
                    } else {
                        esc = false;
                    }
                    parser_pos += 1;
                    if (parser_pos >= input.length) {
                        break;
                    }
                }

            }

            parser_pos += 1;

            resulting_string = sep + resulting_string + sep;

            if (sep == '/') {
                // regexps may have modifiers /regexp/MOD , so fetch those, too
                while (parser_pos < input.length && in_array(input.charAt(parser_pos), wordchar)) {
                    resulting_string += input.charAt(parser_pos);
                    parser_pos += 1;
                }
            }
            return [resulting_string, 'TK_STRING'];
        }

        if (in_array(c, punct)) {
            while (parser_pos < input.length && in_array(c + input.charAt(parser_pos), punct)) {
                c += input.charAt(parser_pos);
                parser_pos += 1;
                if (parser_pos >= input.length) {
                    break;
                }
            }
            return [c, 'TK_OPERATOR'];
        }

        return [c, 'TK_UNKNOWN'];
    }


    //----------------------------------

    indent_string = '';
    while (opt_indent_size--) {
        indent_string += opt_indent_char;
    }

    indent_level = opt_indent_level;

    input = js_source_text;

    last_word = ''; // last 'TK_WORD' passed
    last_type = 'TK_START_EXPR'; // last token type
    last_text = ''; // last token text
    output = [];

    do_block_just_closed = false;
    var_line = false;         // currently drawing var .... ;
    var_line_tainted = false; // false: var a = 5; true: var a = 5, b = 6

    whitespace = "\n\r\t ".split('');
    wordchar = 'abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789_$'.split('');
    punct = '+ - * / % & ++ -- = += -= *= /= %= == === != !== > < >= <= >> << >>> >>>= >>= <<= && &= | || ! !! , : ? ^ ^= |= ::'.split(' ');

    // words which should always start on new line.
    line_starters = 'continue,try,throw,return,var,if,switch,case,default,for,while,break,function'.split(',');

    // states showing if we are currently in expression (i.e. "if" case) - 'EXPRESSION', or in usual block (like, procedure), 'BLOCK'.
    // some formatting depends on that.
    current_mode = 'BLOCK';
    modes = [current_mode];

    parser_pos = 0;
    in_case = false; // flag for parser that case/default has been processed, and next colon needs special attention
    while (true) {
        var t = get_next_token(parser_pos);
        token_text = t[0];
        token_type = t[1];
        if (token_type === 'TK_EOF') {
            break;
        }

        switch (token_type) {

        case 'TK_START_EXPR':
            var_line = false;
            set_mode('EXPRESSION');
            if (last_type === 'TK_END_EXPR' || last_type === 'TK_START_EXPR') {
                // do nothing on (( and )( and ][ and ]( ..
            } else if (last_type !== 'TK_WORD' && last_type !== 'TK_OPERATOR') {
                print_space();
            } else if (in_array(last_word, line_starters) && last_word !== 'function') {
                print_space();
            }
            print_token();
            break;

        case 'TK_END_EXPR':
            print_token();
            restore_mode();
            break;

        case 'TK_START_BLOCK':

            if (last_word === 'do') {
                set_mode('DO_BLOCK');
            } else {
                set_mode('BLOCK');
            }
            if (last_type !== 'TK_OPERATOR' && last_type !== 'TK_START_EXPR') {
                if (last_type === 'TK_START_BLOCK') {
                    print_newline();
                } else {
                    print_space();
                }
            }
            print_token();
            indent();
            break;

        case 'TK_END_BLOCK':
            if (last_type === 'TK_START_BLOCK') {
                // nothing
                trim_output();
                unindent();
            } else {
                unindent();
                print_newline();
            }
            print_token();
            restore_mode();
            break;

        case 'TK_WORD':

            if (do_block_just_closed) {
                print_space();
                print_token();
                print_space();
                break;
            }

            if (token_text === 'case' || token_text === 'default') {
                if (last_text === ':') {
                    // switch cases following one another
                    remove_indent();
                } else {
                    // case statement starts in the same line where switch
                    unindent();
                    print_newline();
                    indent();
                }
                print_token();
                in_case = true;
                break;
            }

            prefix = 'NONE';
            if (last_type === 'TK_END_BLOCK') {
                if (!in_array(token_text.toLowerCase(), ['else', 'catch', 'finally'])) {
                    prefix = 'NEWLINE';
                } else {
                    prefix = 'SPACE';
                    print_space();
                }
            } else if (last_type === 'TK_SEMICOLON' && (current_mode === 'BLOCK' || current_mode === 'DO_BLOCK')) {
                prefix = 'NEWLINE';
            } else if (last_type === 'TK_SEMICOLON' && current_mode === 'EXPRESSION') {
                prefix = 'SPACE';
            } else if (last_type === 'TK_STRING') {
                prefix = 'NEWLINE';
            } else if (last_type === 'TK_WORD') {
                prefix = 'SPACE';
            } else if (last_type === 'TK_START_BLOCK') {
                prefix = 'NEWLINE';
            } else if (last_type === 'TK_END_EXPR') {
                print_space();
                prefix = 'NEWLINE';
            }

            if (last_type !== 'TK_END_BLOCK' && in_array(token_text.toLowerCase(), ['else', 'catch', 'finally'])) {
                print_newline();
            } else if (in_array(token_text, line_starters) || prefix === 'NEWLINE') {
                if (last_text === 'else') {
                    // no need to force newline on else break
                    print_space();
                } else if ((last_type === 'TK_START_EXPR' || last_text === '=') && token_text === 'function') {
                    // no need to force newline on 'function': (function
                    // DONOTHING
                } else if (last_type === 'TK_WORD' && (last_text === 'return' || last_text === 'throw')) {
                    // no newline between 'return nnn'
                    print_space();
                } else if (last_type !== 'TK_END_EXPR') {
                    if ((last_type !== 'TK_START_EXPR' || token_text !== 'var') && last_text !== ':') {
                        // no need to force newline on 'var': for (var x = 0...)
                        if (token_text === 'if' && last_type === 'TK_WORD' && last_word === 'else') {
                            // no newline for } else if {
                            print_space();
                        } else {
                            print_newline();
                        }
                    }
                } else {
                    if (in_array(token_text, line_starters) && last_text !== ')') {
                        print_newline();
                    }
                }
            } else if (prefix === 'SPACE') {
                print_space();
            }
            print_token();
            last_word = token_text;

            if (token_text === 'var') {
                var_line = true;
                var_line_tainted = false;
            }

            if (token_text === 'if' || token_text === 'else') {
                if_line_flag = true;
            }

            break;

        case 'TK_SEMICOLON':

            print_token();
            var_line = false;
            break;

        case 'TK_STRING':

            if (last_type === 'TK_START_BLOCK' || last_type === 'TK_END_BLOCK' || last_type == 'TK_SEMICOLON') {
                print_newline();
            } else if (last_type === 'TK_WORD') {
                print_space();
            }
            print_token();
            break;

        case 'TK_OPERATOR':

            var start_delim = true;
            var end_delim = true;
            if (var_line && token_text !== ',') {
                var_line_tainted = true;
                if (token_text === ':') {
                    var_line = false;
                }
            }

            if (token_text === ':' && in_case) {
                print_token(); // colon really asks for separate treatment
                print_newline();
                break;
            }

            if (token_text === '::') {
                // no spaces around exotic namespacing syntax operator
                print_token();
                break;
            }

            in_case = false;

            if (token_text === ',') {
                if (var_line) {
                    if (var_line_tainted) {
                        print_token();
                        print_newline();
                        var_line_tainted = false;
                    } else {
                        print_token();
                        print_space();
                    }
                } else if (last_type === 'TK_END_BLOCK') {
                    print_token();
                    print_newline();
                } else {
                    if (current_mode === 'BLOCK') {
                        print_token();
                        print_newline();
                    } else {
                        // EXPR od DO_BLOCK
                        print_token();
                        print_space();
                    }
                }
                break;
            } else if (token_text === '--' || token_text === '++') { // unary operators special case
                if (last_text === ';') {
                    // space for (;; ++i)
                    start_delim = true;
                    end_delim = false;
                } else {
                    start_delim = false;
                    end_delim = false;
                }
            } else if (token_text === '!' && last_type === 'TK_START_EXPR') {
                // special case handling: if (!a)
                start_delim = false;
                end_delim = false;
            } else if (last_type === 'TK_OPERATOR') {
                start_delim = false;
                end_delim = false;
            } else if (last_type === 'TK_END_EXPR') {
                start_delim = true;
                end_delim = true;
            } else if (token_text === '.') {
                // decimal digits or object.property
                start_delim = false;
                end_delim = false;

            } else if (token_text === ':') {
                // zz: xx
                // can't differentiate ternary op, so for now it's a ? b: c; without space before colon
                if (last_text.match(/^\d+$/)) {
                    // a little help for ternary a ? 1 : 0;
                    start_delim = true;
                } else {
                    start_delim = false;
                }
            }
            if (start_delim) {
                print_space();
            }

            print_token();

            if (end_delim) {
                print_space();
            }
            break;

        case 'TK_BLOCK_COMMENT':

            print_newline();
            print_token();
            print_newline();
            break;

        case 'TK_COMMENT':

            // print_newline();
            print_space();
            print_token();
            print_newline();
            break;

        case 'TK_UNKNOWN':
            print_token();
            break;
        }

        last_type = token_type;
        last_text = token_text;
    }

    return output.join('');

}


//+++++++++++++++++++++++++++test+++++++++++++++++=

function lazy_escape(str)
{
    return str.replace(/</g, '<').replace(/\>/g, '>').replace(/\n/g, '<br />');
}

function bt(input, expected)
{
    expected = expected || input;

    result = js_beautify(input, {indent_size:indent_size, indent_char:indent_char, preserve_newlines:preserve_newlines});

    if (result != expected) {
        test_result +=
            '\n---- input --------\n' + lazy_escape(input) +
            '\n---- expected -----\n' + lazy_escape(expected) +
            '\n---- received -----\n' + lazy_escape(result) +
            '\n-------------------';
        tests_failed += 1;
    } else {
        tests_passed += 1;
    }

}

function results()
{
    if (tests_failed == 0) {
        test_result += 'All ' + tests_passed + ' tests passed.';
    } else {
        test_result += '\n' + tests_failed + ' tests failed.';
    }
    return test_result;
}

function test_js_beautify()
{

    bt('');
    bt('a        =          1', 'a = 1');
    bt('a=1', 'a = 1');
    bt("a();\n\nb();", "a();\n\nb();");
    bt('var a = 1 var b = 2', "var a = 1\nvar b = 2");
    bt('a = " 12345 "');
    bt("a = ' 12345 '");
    bt('if (a == 1) b = 2', "if (a == 1) b = 2");
    bt('if(1){2}else{3}', "if (1) {\n    2\n} else {\n    3\n}");
    bt('if(1||2)', 'if (1 || 2)');
    bt('(a==1)||(b==2)', '(a == 1) || (b == 2)');
    bt('var a = 1 if (2) 3', "var a = 1\nif (2) 3");
    bt('a = a + 1');
    bt('a = a == 1');
    bt('/12345[^678]*9+/.match(a)');
    bt('a /= 5');
    bt('a = 0.5 * 3');
    bt('a *= 10.55');
    bt('a < .5');
    bt('a <= .5');
    bt('a<.5', 'a < .5');
    bt('a<=.5', 'a <= .5');
    bt('a = 0xff;');
    bt('a=0xff+4', 'a = 0xff + 4');
    bt('a = [1, 2, 3, 4]');
    bt('F*(g/=f)*g+b', 'F * (g /= f) * g + b');
    bt('a.b({c:d})', "a.b({\n    c: d\n})");
    bt('a.b\n(\n{\nc:\nd\n}\n)', "a.b({\n    c: d\n})");
    bt('a=!b', 'a = !b');
    bt('a?b:c', 'a ? b: c'); // 'a ? b : c' would need too make parser more complex to differentiate between ternary op and object assignment
    bt('a?1:2', 'a ? 1 : 2'); // 'a ? b : c' would need too make parser more complex to differentiate between ternary op and object assignment
    bt('a?(b):c', 'a ? (b) : c'); // this works, though
    bt('function void(void) {}');
    bt('if(!a)', 'if (!a)');
    bt('a=~a', 'a = ~a');
    bt('a;/*comment*/b;', "a;\n/*comment*/\nb;");
    bt('if(a)break', "if (a) break");
    bt('if(a){break}', "if (a) {\n    break\n}");
    bt('if((a))', 'if ((a))');
    bt('for(var i=0;;)', 'for (var i = 0;;)');
    bt('a++;', 'a++;');
    bt('for(;;i++)', 'for (;; i++)');
    bt('for(;;++i)', 'for (;; ++i)');
    bt('return(1)', 'return (1)');
    bt('try{a();}catch(b){c();}finally{d();}', "try {\n    a();\n} catch(b) {\n    c();\n} finally {\n    d();\n}");
    bt('(xx)()'); // magic function call
    bt('a[1]()'); // another magic function call
    bt('if(a){b();}else if(', "if (a) {\n    b();\n} else if (");
    bt('switch(x) {case 0: case 1: a(); break; default: break}', "switch (x) {\ncase 0:\ncase 1:\n    a();\n    break;\ndefault:\n    break\n}");
    bt('a !== b');
    bt('if (a) b(); else c();', "if (a) b();\nelse c();");
    bt("// comment\n(function()"); // typical greasemonkey start
    bt("// comment\n(function something()"); // typical greasemonkey start
    bt("{\n\n    x();\n\n}"); // was: duplicating newlines
    bt('if (a in b)');
    //bt('var a, b');
    bt('{a:1, b:2}', "{\n    a: 1,\n    b: 2\n}");
    bt('var l = {\'a\':\'1\', \'b\':\'2\'}', "var l = {\n    'a': '1',\n    'b': '2'\n}");
    bt('if (template.user[n] in bk)');
    bt('{{}/z/}', "{\n    {}\n    /z/\n}");
    bt('return 45', "return 45");
    bt('If[1]', "If[1]");
    bt('Then[1]', "Then[1]");
    bt('a = 1e10', "a = 1e10");
    bt('a = 1.3e10', "a = 1.3e10");
    bt('a = 1.3e-10', "a = 1.3e-10");
    bt('a = -1.3e-10', "a = -1.3e-10");
    bt('a = 1e-10', "a = 1e-10");
    bt('a = e - 10', "a = e - 10");
    bt('a = 11-10', "a = 11 - 10");
    bt("a = 1;// comment\n", "a = 1; // comment\n");
    bt("a = 1; // comment\n", "a = 1; // comment\n");
    bt("a = 1;\n // comment\n", "a = 1;\n// comment\n");

    bt("if (a) {\n    do();\n}"); // was: extra space appended
    bt("if\n(a)\nb()", "if (a) b()"); // test for proper newline removal

    bt("if (a) {\n// comment\n}else{\n// comment\n}", "if (a) {\n    // comment\n} else {\n    // comment\n}"); // if/else statement with empty body
    bt("if (a) {\n// comment\n// comment\n}", "if (a) {\n    // comment\n    // comment\n}"); // multiple comments indentation
    bt("if (a) b() else c()", "if (a) b()\nelse c()");
    bt("if (a) b() else if c() d()", "if (a) b()\nelse if c() d()");

    bt("{}");
    bt("{\n\n}");
    bt("do { a(); } while ( 1 );", "do {\n    a();\n} while ( 1 );");
    bt("do {} while ( 1 );");
    bt("do {\n} while ( 1 );", "do {} while ( 1 );");
    bt("do {\n\n} while ( 1 );");
    bt("var a, b, c, d = 0, c = function() {}, d = '';", "var a, b, c, d = 0,\nc = function() {},\nd = '';");
    bt("var a = x(a, b, c)");
    bt("delete x if (a) b();", "delete x\nif (a) b();");
    bt("delete x[x] if (a) b();", "delete x[x]\nif (a) b();");

    bt("a = 'a'\nb = 'b'");
    bt("a = /reg/exp");
    bt("a = /reg/");
    bt("x(); /reg/exp.match(something)", "x();\n/reg/exp.match(something)");

    bt("function namespace::something()");

    indent_size = 1;
    indent_char = ' ';
    bt('{ one_char() }', "{\n one_char()\n}");

    indent_size = 4;
    indent_char = ' ';
    bt('{ one_char() }', "{\n    one_char()\n}");

    indent_size = 1;
    indent_char = "\t";
    bt('{ one_char() }', "{\n\tone_char()\n}");

    preserve_newlines = false;
    bt('var\na=dont_preserve_newlines', 'var a = dont_preserve_newlines');

    preserve_newlines = true;
    bt('var\na=do_preserve_newlines', 'var\na = do_preserve_newlines');

    return results();
}


//+++++++++++++++++++++++++++++
/*

 Style HTML
---------------


  Written by Nochum Sossonko, (aaa@qq.com)
    $Date$
    $Revision$

  Based on code initially developed by: Einars "elfz" Lielmanis, <aaa@qq.com>
      http://elfz.laacz.lv/beautify/


  You are free to use this in any way you want, in case you find this useful or working for you.

  Usage:
    style_html(html_source);

*/

function style_html(html_source, indent_size, indent_character, max_char) {
//Wrapper function to invoke all the necessary constructors and deal with the output.

  var Parser, multi_parser;

  function Parser() {

    this.pos = 0; //Parser position
    this.token = '';
    this.current_mode = 'CONTENT'; //reflects the current Parser mode: TAG/CONTENT
    this.tags = { //An object to hold tags, their position, and their parent-tags, initiated with default values
      parent: 'parent1',
      parentcount: 1,
      parent1: ''
    };
    this.tag_type = '';
    this.token_text = this.last_token = this.last_text = this.token_type = '';


    this.Utils = { //Uilities made available to the various functions
      whitespace: "\n\r\t ".split(''),
      single_token: 'br,input,link,meta,!doctype,basefont,base,area,hr,wbr,param,img,isindex,?xml,embed'.split(','), //all the single tags for HTML
      extra_liners: 'head,body,/html'.split(','), //for tags that need a line of whitespace before them
      in_array: function (what, arr) {
        for (var i=0; i<arr.length; i++) {
          if (what === arr[i]) {
            return true;
          }
        }
        return false;
      }
    }

    this.get_content = function () { //function to capture regular content between tags

      var char = '';
      var content = [];
      var space = false; //if a space is needed
      while (this.input.charAt(this.pos) !== '<') {
        if (this.pos >= this.input.length) {
          return content.length?content.join(''):['', 'TK_EOF'];
        }

        char = this.input.charAt(this.pos);
        this.pos++;
        this.line_char_count++;


        if (this.Utils.in_array(char, this.Utils.whitespace)) {
          if (content.length) {
            space = true;
          }
          this.line_char_count--;
          continue; //don't want to insert unnecessary space
        }
        else if (space) {
          if (this.line_char_count >= this.max_char) { //insert a line when the max_char is reached
            content.push('\n');
            for (var i=0; i<this.indent_level; i++) {
              content.push(this.indent_string);
            }
            this.line_char_count = 0;
          }
          else{
            content.push(' ');
            this.line_char_count++;
          }
          space = false;
        }
        content.push(char); //letter at-a-time (or string) inserted to an array
      }
      return content.length?content.join(''):'';
    }

    this.get_script = function () { //get the full content of a script to pass to js_beautify

      var char = '';
      var content = [];
      var reg_match = new RegExp('\<\/script' + '\>', 'igm');
      reg_match.lastIndex = this.pos;
      var reg_array = reg_match.exec(this.input);
      var end_script = reg_array?reg_array.index:this.input.length; //absolute end of script
      while(this.pos < end_script) { //get everything in between the script tags
        if (this.pos >= this.input.length) {
          return content.length?content.join(''):['', 'TK_EOF'];
        }

        char = this.input.charAt(this.pos);
        this.pos++;


        content.push(char);
      }
      return content.length?content.join(''):''; //we might not have any content at all
    }

    this.record_tag = function (tag){ //function to record a tag and its parent in this.tags Object
      if (this.tags[tag + 'count']) { //check for the existence of this tag type
        this.tags[tag + 'count']++;
        this.tags[tag + this.tags[tag + 'count']] = this.indent_level; //and record the present indent level
      }
      else { //otherwise initialize this tag type
        this.tags[tag + 'count'] = 1;
        this.tags[tag + this.tags[tag + 'count']] = this.indent_level; //and record the present indent level
      }
      this.tags[tag + this.tags[tag + 'count'] + 'parent'] = this.tags.parent; //set the parent (i.e. in the case of a div this.tags.div1parent)
      this.tags.parent = tag + this.tags[tag + 'count']; //and make this the current parent (i.e. in the case of a div 'div1')
    }

    this.retrieve_tag = function (tag) { //function to retrieve the opening tag to the corresponding closer
      if (this.tags[tag + 'count']) { //if the openener is not in the Object we ignore it
        var temp_parent = this.tags.parent; //check to see if it's a closable tag.
        while (temp_parent) { //till we reach '' (the initial value);
          if (tag + this.tags[tag + 'count'] === temp_parent) { //if this is it use it
            break;
          }
          temp_parent = this.tags[temp_parent + 'parent']; //otherwise keep on climbing up the DOM Tree
        }
        if (temp_parent) { //if we caught something
          this.indent_level = this.tags[tag + this.tags[tag + 'count']]; //set the indent_level accordingly
          this.tags.parent = this.tags[temp_parent + 'parent']; //and set the current parent
        }
        delete this.tags[tag + this.tags[tag + 'count'] + 'parent']; //delete the closed tags parent reference...
        delete this.tags[tag + this.tags[tag + 'count']]; //...and the tag itself
        if (this.tags[tag + 'count'] == 1) {
          delete this.tags[tag + 'count'];
        }
        else {
          this.tags[tag + 'count']--;
        }
      }
    }

    this.get_tag = function () { //function to get a full tag and parse its type
      var char = '';
      var content = [];
      var space = false;

      do {
        if (this.pos >= this.input.length) {
          return content.length?content.join(''):['', 'TK_EOF'];
        }

        char = this.input.charAt(this.pos);
        this.pos++;
        this.line_char_count++;

        if (this.Utils.in_array(char, this.Utils.whitespace)) { //don't want to insert unnecessary space
          space = true;
          this.line_char_count--;
          continue;
        }

        if (char === "'" || char === '"') {
          if (!content[1] || content[1] !== '!') { //if we're in a comment strings don't get treated specially
            char += this.get_unformatted(char);
            space = true;
          }
        }

        if (char === '=') { //no space before =
          space = false;
        }

        if (content.length && content[content.length-1] !== '=' && char !== '>'
            && space) { //no space after = or before >
          if (this.line_char_count >= this.max_char) {
            this.print_newline(false, content);
            this.line_char_count = 0;
          }
          else {
            content.push(' ');
            this.line_char_count++;
          }
          space = false;
        }
        content.push(char); //inserts character at-a-time (or string)
      } while (char !== '>');

      var tag_complete = content.join('');
      var tag_index;
      if (tag_complete.indexOf(' ') != -1) { //if there's whitespace, thats where the tag name ends
        tag_index = tag_complete.indexOf(' ');
      }
      else { //otherwise go with the tag ending
        tag_index = tag_complete.indexOf('>');
      }
      var tag_check = tag_complete.substring(1, tag_index).toLowerCase();
      if (tag_complete.charAt(tag_complete.length-2) === '/' ||
          this.Utils.in_array(tag_check, this.Utils.single_token)) { //if this tag name is a single tag type (either in the list or has a closing /)
        this.tag_type = 'SINGLE';
      }
      else if (tag_check === 'script') { //for later script handling
        this.record_tag(tag_check);
        this.tag_type = 'SCRIPT';
      }
      else if (tag_check === 'style') { //for future style handling (for now it justs uses get_content)
        this.record_tag(tag_check);
        this.tag_type = 'STYLE';
      }
      else if (tag_check.charAt(0) === '!') { //peek for <!-- comment
        if (tag_check.indexOf('[if') != -1) { //peek for <!--[if conditional comment
          if (tag_complete.indexOf('!IE') != -1) { //this type needs a closing --> so...
            var comment = this.get_unformatted('-->', tag_complete); //...delegate to get_unformatted
            content.push(comment);
          }
          this.tag_type = 'START';
        }
        else if (tag_check.indexOf('[endif') != -1) {//peek for <!--[endif end conditional comment
          this.tag_type = 'END';
          this.unindent();
        }
        else if (tag_check.indexOf('[cdata[') != -1) { //if it's a <[cdata[ comment...
          var comment = this.get_unformatted(']]>', tag_complete); //...delegate to get_unformatted function
          content.push(comment);
          this.tag_type = 'SINGLE'; //<![CDATA[ comments are treated like single tags
        }
        else {
          var comment = this.get_unformatted('-->', tag_complete);
          content.push(comment);
          this.tag_type = 'SINGLE';
        }
      }
      else {
        if (tag_check.charAt(0) === '/') { //this tag is a double tag so check for tag-ending
          this.retrieve_tag(tag_check.substring(1)); //remove it and all ancestors
          this.tag_type = 'END';
        }
        else { //otherwise it's a start-tag
          this.record_tag(tag_check); //push it on the tag stack
          this.tag_type = 'START';
        }
        if (this.Utils.in_array(tag_check, this.Utils.extra_liners)) { //check if this double needs an extra line
          this.print_newline(true, this.output);
        }
      }
      return content.join(''); //returns fully formatted tag
    }

    this.get_unformatted = function (delimiter, orig_tag) { //function to return unformatted content in its entirety

      if (orig_tag && orig_tag.indexOf(delimiter) != -1) {
        return '';
      }
      var char = '';
      var content = '';
      var space = true;
      do {


        char = this.input.charAt(this.pos);
        this.pos++

        if (this.Utils.in_array(char, this.Utils.whitespace)) {
          if (!space) {
            this.line_char_count--;
            continue;
          }
          if (char === '\n' || char === '\r') {
            content += '\n';
            for (var i=0; i<this.indent_level; i++) {
              content += this.indent_string;
            }
            space = false; //...and make sure other indentation is erased
            this.line_char_count = 0;
            continue;
          }
        }
        content += char;
        this.line_char_count++;
        space = true;


      } while (content.indexOf(delimiter) == -1);
      return content;
    }

    this.get_token = function () { //initial handler for token-retrieval
      var token;

      if (this.last_token === 'TK_TAG_SCRIPT') { //check if we need to format javascript
        var temp_token = this.get_script();
        if (typeof temp_token !== 'string') {
          return temp_token;
        }
        token = js_beautify(temp_token,
                {indent_size: this.indent_size, indent_char: this.indent_character, indent_level: this.indent_level}); //call the JS Beautifier
        return [token, 'TK_CONTENT'];
      }
      if (this.current_mode === 'CONTENT') {
        token = this.get_content();
        if (typeof token !== 'string') {
          return token;
        }
        else {
          return [token, 'TK_CONTENT'];
        }
      }

      if(this.current_mode === 'TAG') {
        token = this.get_tag();
        if (typeof token !== 'string') {
          return token;
        }
        else {
          var tag_name_type = 'TK_TAG_' + this.tag_type;
          return [token, tag_name_type];
        }
      }
    }

    this.printer = function (js_source, indent_character, indent_size, max_char) { //handles input/output and some other printing functions

      this.input = js_source || ''; //gets the input for the Parser
      this.output = [];
      this.indent_character = indent_character || ' ';
      this.indent_string = '';
      this.indent_size = indent_size || 2;
      this.indent_level = 0;
      this.max_char = max_char || 70; //maximum amount of characters per line
      this.line_char_count = 0; //count to see if max_char was exceeded

      for (var i=0; i<this.indent_size; i++) {
        this.indent_string += this.indent_character;
      }

      this.print_newline = function (ignore, arr) {
        this.line_char_count = 0;
        if (!arr || !arr.length) {
          return;
        }
        if (!ignore) { //we might want the extra line
          while (this.Utils.in_array(arr[arr.length-1], this.Utils.whitespace)) {
            arr.pop();
          }
        }
        arr.push('\n');
        for (var i=0; i<this.indent_level; i++) {
          arr.push(this.indent_string);
        }
      }


      this.print_token = function (text) {
        this.output.push(text);
      }

      this.indent = function () {
        this.indent_level++;
      }

      this.unindent = function () {
        if (this.indent_level > 0) {
          this.indent_level--;
        }
      }
    }
    return this;
  }

  /*_____________________--------------------_____________________*/



  multi_parser = new Parser(); //wrapping functions Parser
  multi_parser.printer(html_source, indent_character, indent_size); //initialize starting values



  while (true) {
      var t = multi_parser.get_token();
      multi_parser.token_text = t[0];
      multi_parser.token_type = t[1];

    if (multi_parser.token_type === 'TK_EOF') {
      break;
    }


    switch (multi_parser.token_type) {
      case 'TK_TAG_START': case 'TK_TAG_SCRIPT': case 'TK_TAG_STYLE':
        multi_parser.print_newline(false, multi_parser.output);
        multi_parser.print_token(multi_parser.token_text);
        multi_parser.indent();
        multi_parser.current_mode = 'CONTENT';
        break;
      case 'TK_TAG_END':
        multi_parser.print_newline(true, multi_parser.output);
        multi_parser.print_token(multi_parser.token_text);
        multi_parser.current_mode = 'CONTENT';
        break;
      case 'TK_TAG_SINGLE':
        multi_parser.print_newline(false, multi_parser.output);
        multi_parser.print_token(multi_parser.token_text);
        multi_parser.current_mode = 'CONTENT';
        break;
      case 'TK_CONTENT':
        if (multi_parser.token_text !== '') {
          multi_parser.print_newline(false, multi_parser.output);
          multi_parser.print_token(multi_parser.token_text);
        }
        multi_parser.current_mode = 'TAG';
        break;
    }
    multi_parser.last_token = multi_parser.token_type;
    multi_parser.last_text = multi_parser.token_text;
  }
  return multi_parser.output.join('');
}