UglifyJS – a JavaScript parser/compressor/beautifier

UglifyJS — a JavaScript parser/compressor/beautifier

This package implements a general-purpose JavaScript parser/compressor/beautifier toolkit. It is developed on NodeJS. With minimal changes it should work on any JavaScript platform (what is Node-specific is usage of exports and JSON.stringify, although the latter is quite portable now).

The tokenizer/parser generates an abstract syntax tree from JS code. You can then traverse the AST to learn more about the code, or do various manipulations on it. This part is implemented in parse-js.js and it’s a port to JavaScript of the excellent parse-js Common Lisp library from Marijn Haverbeke.

The second part of this package, implemented in process.js, inspects and manipulates the AST generated by the parser to provide the following:

ability to re-generate JavaScript code from the AST. Optionally indented—you can use this if you want to “beautify” a program that has been compressed, so that you can inspect the source. But you can also run our code generator to print out an AST without any whitespace, so you achieve compression as well.
shorten variable names (usually to single characters). Our mangler will analyze the code and generate proper variable names, depending on scope and usage, and is smart enough to deal with globals defined elsewhere, or with eval() calls or with{} statements. In short, if eval() or with{} are used in some scope, then all variables in that scope and any variables in the parent scopes will remain unmangled, and any references to such variables remain unmangled as well.
various small optimizations that may lead to faster code but certainly lead to smaller code. Where possible, we do the following:
- foo[“bar”] ==> foo.bar
- remove block brackets {}
- join consecutive var declarations: var a = 10; var b = 20; ==> var a=10,b=20;
- resolve simple constant expressions: 1 +2 * 3 ==> 7. We only do the replacement if the result occupies less bytes; for example 1/3 would translate to 0.333333333333, so in this case we don’t replace it.
- consecutive statements in blocks are merged into a sequence; in many cases, this leaves blocks with a single statement, so then we can remove the block brackets.
- various optimizations for IF statements:
  - if (foo) bar(); else baz(); ==> foo?bar():baz();
  - if (!foo) bar(); else baz(); ==> foo?baz():bar();
  - if (foo) bar(); ==> foo&&bar();
  - if (!foo) bar(); ==> foo||bar();
  - if (foo) return bar(); else return baz(); ==> return foo?bar():baz();
  - if (foo) return bar(); else something(); ==> {if(foo)return bar();something()}

Usage

There is a helper script now — bin/uglifyjs — that uses the library to compress a script using the maximum compression settings. Synopsis:

uglifyjs [ options... ] [ filename ]

filename should be the last argument and should name the file from which to read the JavaScript code. If you don’t specify it, it will read code from STDIN.

Supported options:

-b or --beautify — output indented code; when passed, additional options control the beautifier:
- -i N or --indent N — indentation level (number of spaces)
- -q or --quote-keys — quote keys in literal objects (by default, only keys that cannot be identifier names will be quotes).
-nm or --no-mangle — don’t mangle variable names
-ns or --no-squeeze — don’t call ast_squeeze() (which does various optimizations that result in smaller, less readable code).
-mt or --mangle-toplevel — mangle names in the toplevel scope too (by default we don’t do this).
--no-seqs — when ast_squeeze() is called (thus, unless you pass --no-squeeze) it will reduce consecutive statements in blocks into a sequence. For example, “a = 10; b = 20; foo();” will be written as “a=10,b=20,foo();”. In various occasions, this allows us to discard the block brackets (since the block becomes a single statement). This is ON by default because it seems safe and saves a few hundred bytes on some libs that I tested it on, but pass --no-seqs to disable it.
-nc or --no-copyright — by default, uglifyjs will keep the initial comment tokens in the generated code (assumed to be copyright information etc.). If you pass this it will discard it.
-o filename or --output filename — put the result in filename. If this isn’t given, the result goes to standard output (or see next one).
--overwrite — if the code is read from a file (not from STDIN) and you pass --overwrite then the output will be written in the same file.
--ast — pass this if you want to get the Abstract Syntax Tree instead of JavaScript as output. Useful for debugging or learning more about the internals.
-v or --verbose — output some notes on STDERR (for now just how long each operation takes).

API

Symlink the lib directory as ~/.node\_libraries/uglifyjs, so that the require calls in the following sample will work:

var jsp = require("uglifyjs/parse-js");
var pro = require("uglifyjs/process");

var orig_code = "... JS code here";
var ast = jsp.parse(orig_code); // parse code and get the initial AST
ast = pro.ast_mangle(ast); // get a new AST with mangled names
ast = pro.ast_squeeze(ast); // get an AST with compression optimizations
var final_code = pro.gen_code(ast); // compressed code here

The above performs the full compression that is possible right now. As you can see, there are a sequence of steps which you can apply. For example if you want compressed output but for some reason you don’t want to mangle variable names, you would simply skip the line that calls pro.ast_mangle(ast).

Some of these functions take optional arguments. Here’s a description:

jsp.parse(code, strict_semicolons) – parses JS code and returns an AST. strict_semicolons is optional and defaults to false. If you pass true then the parser will throw an error when it expects a semicolon and it doesn’t find it. For most JS code you don’t want that, but it’s useful if you want to strictly sanitize your code.
pro.ast_mangle(ast, do_toplevel) – generates a new AST containing mangled (compressed) variable and function names. By default it doesn’t touch the names defined in the toplevel scope, but if you pass true as second argument it will compress them as well.
pro.ast_squeeze(ast, options) – employs further optimizations designed to reduce the size of the code that gen_code would generate from the AST. Returns a new AST. options can be a hash; for now the only supported option is make\_seqs (default true) which will cause consecutive statements in a block to be merged using the “sequence” (comma) operator.
pro.gen_code(ast, beautify) – generates JS code from the AST. By default it’s minified, but if you pass true for the second argument it will be nicely formatted and indented. Additionally, you can control the behavior by passing a hash for beautify, where the following options are supported (below you can see the default values):
- indent_start: 0 – initial indentation in spaces
- indent_level: 4 – indentation level, in spaces (pass an even number)
- quote_keys: false – if you pass true it will quote all keys in literal objects

Beautifier shortcoming – no more comments

The beautifier can be used as a general purpose indentation tool. It’s useful when you want to make a minified file readable. One limitation, though, is that it discards all comments, so you don’t really want to use it to reformat your code, unless you don’t have, or don’t care about, comments.

In fact it’s not the beautifier who discards comments — they are dumped at the parsing stage, when we build the initial AST. Comments don’t really make sense in the AST, and while we could add nodes for them, it would be inconvenient because we’d have to add special rules to ignore them at all the processing stages.

Compression – how good is it?

There are a few popular JS minifiers nowadays – the two most well known being the GoogleClosure (GCL) compiler and the YUI compressor. For some reason they are both written in Java. I didn’t really hope to beat any of them, but finally I did – UglifyJS compresses better than the YUI compressor, and safer than GoogleClosure.

I tested it on two big libraries. DynarchLIB is my own, and it’s big enough to contain probably all the JavaScript tricks known to mankind. jQuery is definitely the most popular JavaScript library (to some people, it’s a synonym to JavaScript itself).

I cannot swear that there are no bugs in the generated codes, but they appear to work fine.

Compression results:

Library	Orig. size	UglifyJS	YUI	GCL
DynarchLIB	636896	241441	246452 (+5011)	240439 (-1002) (buggy)
jQuery	163855	72006	79702 (+7696)	71858 (-148)

UglifyJS is the fastest to run. On my laptop UglifyJS takes 1.35s for DynarchLIB, while YUI takes 2.7s and GCL takes 6.5s.

GoogleClosure does a lot of smart ass optimizations. I had to strive really hard to get close to it. It should be possible to even beat it, but then again, GCL has a gazillion lines of code and runs terribly slow, so I’m not sure it worths spending the effort to save a few bytes. Also, GCL doesn’t cope with eval() or with{} – it just dumps a warning and proceeds to mangle names anyway; my DynarchLIB compiled with it is buggy because of this.

UglifyJS consists of ~1100 lines of code for the tokenizer/parser, and ~1100 lines for the compressor and code generator. That should make it very maintainable and easily extensible, so I would say it has a good place in this field and it’s bound to become the de-facto standard JS minifier. And I shall rule the world. :-) Use it, and spread the word!

Bugs?

Unfortunately, for the time being there is no automated test suite. But I ran the compressor manually on non-trivial code, and then I tested that the generated code works as expected. A few hundred times.

DynarchLIB was started in times when there was no good JS minifier. Therefore I was quite religious about trying to write short code manually, and as such DL contains a lot of syntactic hacks[1] such as “foo == bar ? a = 10 : b = 20”, though the more readable version would clearly be to use “if/else”.

Since the parser/compressor runs fine on DL and jQuery, I’m quite confident that it’s solid enough for production use. If you can identify any bugs, I’d love to hear about them (use the Google Group or email me directly).

[1] I even reported a few bugs and suggested some fixes in the original parse-js library, and Marijn pushed fixes literally in minutes.

License

UglifyJS is released under a ZLIB-like license:

Copyright 2010 (c) Mihai Bazon <mihai.bazon@gmail.com>
Parser based on parse-js (http://marijn.haverbeke.nl/parse-js/).

This software is provided 'as-is', without any express or implied
warranty. In no event will the authors be held liable for any
damages arising from the use of this software.

Permission is granted to anyone to use this software for any
purpose, including commercial applications, and to alter it and
redistribute it freely, subject to the following restrictions:

1. The origin of this software must not be misrepresented; you must
   not claim that you wrote the original software. If you use this
   software in a product, an acknowledgment in the product
   documentation would be appreciated but is not required.

2. Altered source versions must be plainly marked as such, and must
   not be misrepresented as being the original software.

3. This notice may not be removed or altered from any source
   distribution.

thegreatape / UglifyJS