Skip to content

Instantly share code, notes, and snippets.

@azu

azu/js_Beautifier.ng

Created Jun 25, 2010
Embed
What would you like to do?
クリップボードのJavaScriptコードを整形してクリップボードに返すNILScript
// ==UserScript==
// @name js_Beautifier
// @namespace http://efcl.info/
// @description クリップボードのJavaScriptコードを整形してクリップボードに返す
// @author azu
// @homepage http://efcl.info/
// @twitter https://twitter.com/azu_re
// ==/UserScript==
var Clipboard=require('Clipboard').Clipboard;
var formated = js_beautify(Clipboard.text, {indent_size: 4 , preserve_newlines :false});
Clipboard.text = formated;
/*jslint onevar: false, plusplus: false */
/*
JS Beautifier
---------------
Written by Einar Lielmanis, <einar@jsbeautifier.org>
http://jsbeautifier.org/
Originally converted to javascript by Vital, <vital76@gmail.com>
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,
space_after_anon_function (default false) — if true, then space is added between "function ()"
(jslint is happy about this); if false, then the common "function()" output is used.
braces_on_own_line (default false) - ANSI / Allman brace style, each opening/closing brace gets its own line.
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_last_text, last_word, flags, flag_store, indent_string;
var whitespace, wordchar, punct, parser_pos, line_starters, digits;
var prefix, token_type, do_block_just_closed;
var wanted_newline, just_added_newline, n_newlines;
// Some interpreters have unexpected results with foo = baz || bar;
options = options ? options : {};
var opt_braces_on_own_line = options.braces_on_own_line ? options.braces_on_own_line : false;
var opt_indent_size = options.indent_size ? options.indent_size : 4;
var opt_indent_char = options.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 ? options.indent_level : 0; // starting indentation
var opt_space_after_anon_function = options.space_after_anon_function === 'undefined' ? false : options.space_after_anon_function;
var opt_keep_array_indentation = typeof options.keep_array_indentation === 'undefined' ? true : options.keep_array_indentation;
just_added_newline = false;
// cache the source's length.
var input_length = js_source_text.length;
function trim_output() {
while (output.length && (output[output.length - 1] === ' ' || output[output.length - 1] === indent_string)) {
output.pop();
}
}
function is_array(mode) {
return mode === '[EXPRESSION]' || mode === '[INDENTED-EXPRESSION]';
}
function print_newline(ignore_repeated) {
flags.eat_next_space = false;
if (opt_keep_array_indentation && is_array(flags.mode)) {
return;
}
ignore_repeated = typeof ignore_repeated === 'undefined' ? true : ignore_repeated;
flags.if_line = false;
trim_output();
if (!output.length) {
return; // no newline on start of file
}
if (output[output.length - 1] !== "\n" || !ignore_repeated) {
just_added_newline = true;
output.push("\n");
}
for (var i = 0; i < flags.indentation_level + (flags.var_line && flags.var_line_reindented ? 1 : 0); i += 1) {
output.push(indent_string);
}
}
function print_single_space() {
if (flags.eat_next_space) {
flags.eat_next_space = false;
return;
}
var last_output = ' ';
if (output.length) {
last_output = output[output.length - 1];
}
if (last_output !== ' ' && last_output !== '\n' && last_output !== indent_string) { // prevent occassional duplicate space
output.push(' ');
}
}
function print_token() {
just_added_newline = false;
flags.eat_next_space = false;
output.push(token_text);
}
function indent() {
flags.indentation_level += 1;
}
function remove_indent() {
if (output.length && output[output.length - 1] === indent_string) {
output.pop();
}
}
function set_mode(mode) {
if (flags) {
flag_store.push(flags);
}
flags = {
mode: mode,
var_line: false,
var_line_tainted: false,
var_line_reindented: false,
in_html_comment: false,
if_line: false,
in_case: false,
eat_next_space: false,
indentation_baseline: -1,
indentation_level: (flags ? flags.indentation_level + ((flags.var_line && flags.var_line_reindented) ? 1 : 0) : opt_indent_level)
};
}
function is_expression(mode) {
return mode === '[EXPRESSION]' || mode === '[INDENTED-EXPRESSION]' || mode === '(EXPRESSION)';
}
function restore_mode() {
do_block_just_closed = flags.mode === 'DO_BLOCK';
if (flag_store.length > 0) {
flags = flag_store.pop();
}
}
function in_array(what, arr) {
for (var i = 0; i < arr.length; i += 1) {
if (arr[i] === what) {
return true;
}
}
return false;
}
// Walk backwards from the colon to find a '?' (colon is part of a ternary op)
// or a '{' (colon is part of a class literal). Along the way, keep track of
// the blocks and expressions we pass so we only trigger on those chars in our
// own level, and keep track of the colons so we only trigger on the matching '?'.
function is_ternary_op() {
var level = 0,
colon_count = 0;
for (var i = output.length - 1; i >= 0; i--) {
switch (output[i]) {
case ':':
if (level === 0) {
colon_count++;
}
break;
case '?':
if (level === 0) {
if (colon_count === 0) {
return true;
} else {
colon_count--;
}
}
break;
case '{':
if (level === 0) {
return false;
}
level--;
break;
case '(':
case '[':
level--;
break;
case ')':
case ']':
case '}':
level++;
break;
}
}
}
function get_next_token() {
n_newlines = 0;
if (parser_pos >= input_length) {
return ['', 'TK_EOF'];
}
wanted_newline = false;
var c = input.charAt(parser_pos);
parser_pos += 1;
var keep_whitespace = opt_keep_array_indentation && is_array(flags.mode);
if (keep_whitespace) {
//
// slight mess to allow nice preservation of array indentation and reindent that correctly
// first time when we get to the arrays:
// var a = [
// ....'something'
// we make note of whitespace_count = 4 into flags.indentation_baseline
// so we know that 4 whitespaces in original source match indent_level of reindented source
//
// and afterwards, when we get to
// 'something,
// .......'something else'
// we know that this should be indented to indent_level + (7 - indentation_baseline) spaces
//
var whitespace_count = 0;
while (in_array(c, whitespace)) {
if (c === "\n") {
trim_output();
output.push("\n");
just_added_newline = true;
whitespace_count = 0;
} else {
if (c === '\t') {
whitespace_count += 4;
} else {
whitespace_count += 1;
}
}
if (parser_pos >= input_length) {
return ['', 'TK_EOF'];
}
c = input.charAt(parser_pos);
parser_pos += 1;
}
if (flags.indentation_baseline === -1) {
flags.indentation_baseline = whitespace_count;
}
if (just_added_newline) {
var i;
for (i = 0; i < flags.indentation_level + 1; i += 1) {
output.push(indent_string);
}
if (flags.indentation_baseline !== -1) {
for (i = 0; i < whitespace_count - flags.indentation_baseline; i++) {
output.push(' ');
}
}
}
} else {
while (in_array(c, whitespace)) {
if (c === "\n") {
n_newlines += 1;
}
if (parser_pos >= input_length) {
return ['', 'TK_EOF'];
}
c = input.charAt(parser_pos);
parser_pos += 1;
}
if (opt_preserve_newlines) {
if (n_newlines > 1) {
for (i = 0; i < n_newlines; i += 1) {
print_newline(i === 0);
just_added_newline = true;
}
}
}
wanted_newline = n_newlines > 0;
}
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) === '-' || input.charAt(parser_pos) === '+')) {
var sign = input.charAt(parser_pos);
parser_pos += 1;
var t = get_next_token(parser_pos);
c += sign + t[0];
return [c, 'TK_WORD'];
}
if (c === 'in') { // hack for 'in' operator
return [c, 'TK_OPERATOR'];
}
if (wanted_newline && last_type !== 'TK_OPERATOR' && !flags.if_line && (opt_preserve_newlines || last_text !== 'var')) {
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 /* ... */
var inline_comment = true;
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) {
c = input.charAt(parser_pos);
comment += c;
if (c === '\x0d' || c === '\x0a') {
inline_comment = false;
}
parser_pos += 1;
if (parser_pos >= input_length) {
break;
}
}
}
parser_pos += 2;
if (inline_comment) {
return ['/*' + comment + '*/', 'TK_INLINE_COMMENT'];
} else {
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' && in_array(last_text, ['return', 'do'])) || (last_type === 'TK_START_EXPR' || last_type === 'TK_START_BLOCK' || last_type === 'TK_END_BLOCK' || last_type === 'TK_OPERATOR' || last_type === 'TK_EQUALS' || last_type === 'TK_EOF' || last_type === 'TK_SEMICOLON')))) { // regexp
var sep = c;
var esc = false;
var resulting_string = c;
if (parser_pos < input_length) {
if (sep === '/') {
//
// handle regexp separately...
//
var in_char_class = false;
while (esc || in_char_class || input.charAt(parser_pos) !== sep) {
resulting_string += input.charAt(parser_pos);
if (!esc) {
esc = input.charAt(parser_pos) === '\\';
if (input.charAt(parser_pos) === '[') {
in_char_class = true;
} else if (input.charAt(parser_pos) === ']') {
in_char_class = false;
}
} else {
esc = false;
}
parser_pos += 1;
if (parser_pos >= input_length) {
// incomplete string/rexp when end-of-file reached.
// bail out with what had been received so far.
return [resulting_string, 'TK_STRING'];
}
}
} else {
//
// and handle string also separately
//
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) {
// incomplete string/rexp when end-of-file reached.
// bail out with what had been received so far.
return [resulting_string, 'TK_STRING'];
}
}
}
}
parser_pos += 1;
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 (c === '#') {
// Spidermonkey-specific sharp variables for circular references
// https://developer.mozilla.org/En/Sharp_variables_in_JavaScript
// http://mxr.mozilla.org/mozilla-central/source/js/src/jsscan.cpp around line 1935
var sharp = '#';
if (parser_pos < input_length && in_array(input.charAt(parser_pos), digits)) {
do {
c = input.charAt(parser_pos);
sharp += c;
parser_pos += 1;
} while (parser_pos < input_length && c !== '#' && c !== '=');
if (c === '#') {
//
} else if (input.charAt(parser_pos) === '[' && input.charAt(parser_pos + 1) === ']') {
sharp += '[]';
parser_pos += 2;
} else if (input.charAt(parser_pos) === '{' && input.charAt(parser_pos + 1) === '}') {
sharp += '{}';
parser_pos += 2;
}
return [sharp, 'TK_WORD'];
}
}
if (c === '<' && input.substring(parser_pos - 1, parser_pos + 3) === '<!--') {
parser_pos += 3;
flags.in_html_comment = true;
return ['<!--', 'TK_COMMENT'];
}
if (c === '-' && flags.in_html_comment && input.substring(parser_pos - 1, parser_pos + 2) === '-->') {
flags.in_html_comment = false;
parser_pos += 2;
if (wanted_newline) {
print_newline();
}
return ['-->', 'TK_COMMENT'];
}
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;
}
}
if (c === '=') {
return [c, 'TK_EQUALS'];
} else {
return [c, 'TK_OPERATOR'];
}
}
return [c, 'TK_UNKNOWN'];
}
//----------------------------------
indent_string = '';
while (opt_indent_size > 0) {
indent_string += opt_indent_char;
opt_indent_size -= 1;
}
input = js_source_text;
last_word = ''; // last 'TK_WORD' passed
last_type = 'TK_START_EXPR'; // last token type
last_text = ''; // last token text
last_last_text = ''; // pre-last token text
output = [];
do_block_just_closed = false;
whitespace = "\n\r\t ".split('');
wordchar = 'abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789_$'.split('');
digits = '0123456789'.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.
flag_store = [];
set_mode('BLOCK');
parser_pos = 0;
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':
if (token_text === '[') {
if (last_type === 'TK_WORD' || last_text === ')') {
// this is array index specifier, break immediately
// a[x], fn()[x]
if (in_array(last_text, line_starters)) {
print_single_space();
}
set_mode('(EXPRESSION)');
print_token();
break;
}
if (flags.mode === '[EXPRESSION]' || flags.mode === '[INDENTED-EXPRESSION]') {
if (last_last_text === ']' && last_text === ',') {
// ], [ goes to new line
if (flags.mode === '[EXPRESSION]') {
flags.mode = '[INDENTED-EXPRESSION]';
if (!opt_keep_array_indentation) {
indent();
}
}
set_mode('[EXPRESSION]');
if (!opt_keep_array_indentation) {
print_newline();
}
} else if (last_text === '[') {
if (flags.mode === '[EXPRESSION]') {
flags.mode = '[INDENTED-EXPRESSION]';
if (!opt_keep_array_indentation) {
indent();
}
}
set_mode('[EXPRESSION]');
if (!opt_keep_array_indentation) {
print_newline();
}
} else {
set_mode('[EXPRESSION]');
}
} else {
set_mode('[EXPRESSION]');
}
} else {
set_mode('(EXPRESSION)');
}
if (last_text === ';' || last_type === 'TK_START_BLOCK') {
print_newline();
} else if (last_type === 'TK_END_EXPR' || last_type === 'TK_START_EXPR' || last_type === 'TK_END_BLOCK' || last_text === '.') {
// do nothing on (( and )( and ][ and ]( and .(
} else if (last_type !== 'TK_WORD' && last_type !== 'TK_OPERATOR') {
print_single_space();
} else if (last_word === 'function') {
// function() vs function ()
if (opt_space_after_anon_function) {
print_single_space();
}
} else if (in_array(last_text, line_starters) || last_text === 'catch') {
print_single_space();
}
print_token();
break;
case 'TK_END_EXPR':
if (token_text === ']') {
if (opt_keep_array_indentation) {
if (last_text === '}') {
// trim_output();
// print_newline(true);
remove_indent();
print_token();
restore_mode();
break;
}
} else {
if (flags.mode === '[INDENTED-EXPRESSION]') {
if (last_text === ']') {
restore_mode();
print_newline();
print_token();
break;
}
}
}
}
restore_mode();
print_token();
break;
case 'TK_START_BLOCK':
if (last_word === 'do') {
set_mode('DO_BLOCK');
} else {
set_mode('BLOCK');
}
if (opt_braces_on_own_line) {
if (last_type !== 'TK_OPERATOR') {
if (last_text == 'return') {
print_single_space();
} else {
print_newline(true);
}
}
print_token();
indent();
} else {
if (last_type !== 'TK_OPERATOR' && last_type !== 'TK_START_EXPR') {
if (last_type === 'TK_START_BLOCK') {
print_newline();
} else {
print_single_space();
}
}
indent();
print_token();
}
break;
case 'TK_END_BLOCK':
restore_mode();
if (opt_braces_on_own_line) {
print_newline();
print_token();
} else {
if (last_type === 'TK_START_BLOCK') {
// nothing
if (just_added_newline) {
remove_indent();
} else {
// {}
trim_output();
}
} else {
print_newline();
}
print_token();
}
break;
case 'TK_WORD':
// no, it's not you. even I have problems understanding how this works
// and what does what.
if (do_block_just_closed) {
// do {} ## while ()
print_single_space();
print_token();
print_single_space();
do_block_just_closed = false;
break;
}
if (token_text === 'function') {
if ((just_added_newline || last_text === ';') && last_text !== '{') {
// make sure there is a nice clean space of at least one blank line
// before a new function definition
n_newlines = just_added_newline ? n_newlines : 0;
for (var i = 0; i < 2 - n_newlines; i++) {
print_newline(false);
}
}
}
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
flags.indentation_level--;
print_newline();
flags.indentation_level++;
}
print_token();
flags.in_case = true;
break;
}
prefix = 'NONE';
if (last_type === 'TK_END_BLOCK') {
if (!in_array(token_text.toLowerCase(), ['else', 'catch', 'finally'])) {
prefix = 'NEWLINE';
} else {
if (opt_braces_on_own_line) {
prefix = 'NEWLINE';
} else {
prefix = 'SPACE';
print_single_space();
}
}
} else if (last_type === 'TK_SEMICOLON' && (flags.mode === 'BLOCK' || flags.mode === 'DO_BLOCK')) {
prefix = 'NEWLINE';
} else if (last_type === 'TK_SEMICOLON' && is_expression(flags.mode)) {
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_single_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_single_space();
} else if ((last_type === 'TK_START_EXPR' || last_text === '=' || last_text === ',') && token_text === 'function') {
// no need to force newline on 'function': (function
// DONOTHING
} else if (last_text === 'return' || last_text === 'throw') {
// no newline between 'return nnn'
print_single_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_word === 'else' && last_text !== '{') {
// no newline for } else if {
print_single_space();
} else {
print_newline();
}
}
} else {
if (in_array(token_text, line_starters) && last_text !== ')') {
print_newline();
}
}
} else if (prefix === 'SPACE') {
print_single_space();
}
print_token();
last_word = token_text;
if (token_text === 'var') {
flags.var_line = true;
flags.var_line_reindented = false;
flags.var_line_tainted = false;
}
if (token_text === 'if' || token_text === 'else') {
flags.if_line = true;
}
break;
case 'TK_SEMICOLON':
print_token();
flags.var_line = false;
flags.var_line_reindented = 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_single_space();
}
print_token();
break;
case 'TK_EQUALS':
if (flags.var_line) {
// just got an '=' in a var-line, different formatting/line-breaking, etc will now be done
flags.var_line_tainted = true;
}
print_single_space();
print_token();
print_single_space();
break;
case 'TK_OPERATOR':
var space_before = true;
var space_after = true;
if (flags.var_line && token_text === ',' && (is_expression(flags.mode))) {
// do not break on comma, for(var a = 1, b = 2)
flags.var_line_tainted = false;
}
if (flags.var_line) {
if (token_text === ',') {
if (flags.var_line_tainted) {
print_token();
flags.var_line_reindented = true;
flags.var_line_tainted = false;
print_newline();
break;
} else {
flags.var_line_tainted = false;
}
// } else if (token_text === ':') {
// hmm, when does this happen? tests don't catch this
// flags.var_line = false;
}
}
if (last_text === 'return' || last_text === 'throw') {
// "return" had a special handling in TK_WORD. Now we need to return the favor
print_single_space();
print_token();
break;
}
if (token_text === ':' && flags.in_case) {
print_token(); // colon really asks for separate treatment
print_newline();
flags.in_case = false;
break;
}
if (token_text === '::') {
// no spaces around exotic namespacing syntax operator
print_token();
break;
}
if (token_text === ',') {
if (flags.var_line) {
if (flags.var_line_tainted) {
print_token();
print_newline();
flags.var_line_tainted = false;
} else {
print_token();
print_single_space();
}
} else if (last_type === 'TK_END_BLOCK' && flags.mode !== "(EXPRESSION)") {
print_token();
print_single_space();
} else {
if (flags.mode === 'BLOCK') {
print_token();
print_newline();
} else {
// EXPR or DO_BLOCK
print_token();
print_single_space();
}
}
break;
// } else if (in_array(token_text, ['--', '++', '!']) || (in_array(token_text, ['-', '+']) && (in_array(last_type, ['TK_START_BLOCK', 'TK_START_EXPR', 'TK_EQUALS']) || in_array(last_text, line_starters) || in_array(last_text, ['==', '!=', '+=', '-=', '*=', '/=', '+', '-'])))) {
} else if (in_array(token_text, ['--', '++', '!']) || (in_array(token_text, ['-', '+']) && (in_array(last_type, ['TK_START_BLOCK', 'TK_START_EXPR', 'TK_EQUALS', 'TK_OPERATOR']) || in_array(last_text, line_starters)))) {
// unary operators (and binary +/- pretending to be unary) special cases
space_before = false;
space_after = false;
if (last_text === ';' && is_expression(flags.mode)) {
// for (;; ++i)
// ^^^
space_before = true;
}
if (last_type === 'TK_WORD' && in_array(last_text, line_starters)) {
space_before = true;
}
if (flags.mode === 'BLOCK' && (last_text === '{' || last_text === ';')) {
// { foo; --i }
// foo(); --bar;
print_newline();
}
} else if (token_text === '.') {
// decimal digits or object.property
space_before = false;
} else if (token_text === ':') {
if (!is_ternary_op()) {
space_before = false;
}
}
if (space_before) {
print_single_space();
}
print_token();
if (space_after) {
print_single_space();
}
if (token_text === '!') {
// flags.eat_next_space = true;
}
break;
case 'TK_BLOCK_COMMENT':
var lines = token_text.split(/\x0a|\x0d\x0a/);
if (/^\/\*\*/.test(token_text)) {
// javadoc: reformat and reindent
print_newline();
output.push(lines[0]);
for (i = 1; i < lines.length; i++) {
print_newline();
output.push(' ');
output.push(lines[i].replace(/^\s\s*|\s\s*$/, ''));
}
} else {
// simple block comment: leave intact
if (lines.length > 1) {
// multiline comment block starts with a new line
print_newline();
trim_output();
} else {
// single-line /* comment */ stays where it is
print_single_space();
}
for (i = 0; i < lines.length; i++) {
output.push(lines[i]);
output.push('\n');
}
}
print_newline();
break;
case 'TK_INLINE_COMMENT':
print_single_space();
print_token();
if (is_expression(flags.mode)) {
print_single_space();
} else {
print_newline();
}
break;
case 'TK_COMMENT':
// print_newline();
if (wanted_newline) {
print_newline();
} else {
print_single_space();
}
print_token();
print_newline();
break;
case 'TK_UNKNOWN':
print_token();
break;
}
last_last_text = last_text;
last_type = token_type;
last_text = token_text;
}
return output.join('').replace(/[\n ]+$/, '');
}
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
You can’t perform that action at this time.