JavaScript-Is-Weird as a compressor

Have you seen the "JavaScript Is Weird (EXTREME EDITION)" video?

This one, in case you're wandering "which one of them?".

It's a ~22-minute video that abuses type coercion issues in JavaScript to create a prototype of a JavaScript transpiler that outputs valid code consistinting of only the symbols ( ) = > { } [ + ] ! / -. As an example, this "Hello, world!" program in JavaScript:

console.log("Hello world!");

Gets converted to this other valid JavaScript program, that produces the same result but whose source is really weird. Here's the first few bytes of it:

(()=>{})[({}+[])[+!![] + +!![] + +!![] + +!![] + +!![]]+({}+[])[+!![]]+(
(+!![]/+[])+[])[+!![] + +!![] + +!![] + +!![]]+(![]+[])[+!![] + +!![] + 
+!![]]+({}+[])[+!![] + +!![] + +!![] + +!![] + +!![] + +!![]]+(!![]+[])[
+!![]]+(!![]+[])[+!![] + +!![]]+({}+[])[+!![] + +!![] + +!![] + +!![] + 

If you haven't seen the video yet, go do it - this post will be waiting right here.

So what?

So a few months ago I've watch the video and had a weird idea. If the resulting files are basically a different encoding of the same source code, but using a representation that uses less symbols, is there a chance that GZIP'ing those files would result in smaller files than GZIP'ing the original source files? Exchanging GZIP'ed files over the wire is standard practice for web browsers for a while now, so if the hypothesis (a really weak one, yes) worked, this weird idea may become something useful-ish.

And if this worked, how would the compression compare to JavaScript minifiers? And what about runtime performance? Maybe there were some scenarios in which it's OK to trade some runtime performance for smaller file sizes.

Let's try it

First of all, I've downloaded the video's linked repo with the sample code and run it locally. Because knowing some NodeJS code worked a year says very little about that same code working now - and I even didn't knew if it worked at all.

$ git clone https://github.com/lowbyteproductions/JavaScript-Is-Weird
Cloning into 'JavaScript-Is-Weird'...
remote: Enumerating objects: 4, done.
remote: Counting objects: 100% (3/3), done.
remote: Compressing objects: 100% (3/3), done.
remote: Total 4 (delta 0), reused 0 (delta 0), pack-reused 1
Receiving objects: 100% (4/4), done.
$ cd JavaScript-Is-Weird/
$ node index.js
(()=>{})[({}+[])[+!![] + +!![] + +!![] + +!![] + +!![]]+({}+[])[+!![]]+(
... snip ...
$

Inspecting the code, it had a hard-coded call to the compile function with the console.log("Hello, world!"); code in it. So I quickly changed the program to receive the name of the source file as a command line argument - by asking ChatGPT for that trivial piece of code.

I've ran the program against itself, and checked that it kinda worked - it was producing some output, but I couldn't ran itself once again because the transpiler wraps the source code in an anonymous function call that doesn't support requireing. But, hey - we have some output!

$ node weird.js weird.js > weird.weird.js
$ gzip < weird.weird.js > weird.weird.js.gzip
$ ls -l weird.*
-rw-r--r--  1 mgarcia  staff     2065 Oct 16 01:28 weird.js
-rw-r--r--  1 mgarcia  staff  7702811 Oct 16 02:44 weird.weird.js
-rw-r--r--  1 mgarcia  staff    85691 Oct 16 02:45 weird.weird.js.gzip

(Yes, I've also renamed the script as weird.js and adopted the .weird sub-extension since it seemed appropriate and YOLO.)

OK, so we are kind of testing the hypotesis already - that file doesn't feel smaller than the original one. And let's not even talk about minification or gziping the source file.

$ npx minify weird.js > weird.min.js
$ gzip < weird.js > weird.js.gzip
$ ls -l weird.*
-rw-r--r--  1 mgarcia  staff     2065 Oct 16 01:28 weird.js
-rw-r--r--  1 mgarcia  staff      756 Oct 16 02:49 weird.js.gzip
-rw-r--r--  1 mgarcia  staff     1242 Oct 16 02:48 weird.min.js
-rw-r--r--  1 mgarcia  staff  7702811 Oct 16 02:44 weird.weird.js
-rw-r--r--  1 mgarcia  staff    85691 Oct 16 02:45 weird.weird.js.gzip

So the original file is 2065 bytes long, the weird version is ~7MB, and the weird+GZIP version is 85k long (41.5x the size) - while the minified version is 0.6x the size, and GZIP is 0.36x.

I didn't even care to check the runtime performance, since there's no benefits to trade off.

Does this result apply to other files?

So maybe this one script was weird enough that it didn't compress that well? I had to try with some other files, so I got a few examples of JavaScript files that weren't so long - after some file lenght, an Array.join in the transpiler fails with a RangeException, so no JQuery for our tests.

You can check the files in the sample/ directory. They all show similar, aweful results:

$ ls -l sample/
total 683064
-rw-r--r--  1 mgarcia  staff      19648 Oct 16 01:39 dommy-2.0.js
-rw-r--r--  1 mgarcia  staff       5379 Oct 16 02:57 dommy-2.0.js.gz
-rw-r--r--  1 mgarcia  staff       8762 Oct 16 01:40 dommy-2.0.min.js
-rw-r--r--  1 mgarcia  staff   51015234 Oct 16 01:40 dommy-2.0.weird.js
-rw-r--r--  1 mgarcia  staff     587229 Oct 16 01:42 dommy-2.0.weird.js.gz
-rw-r--r--  1 mgarcia  staff     115023 Oct 16 01:50 lodash-4.17.15.js
-rw-r--r--  1 mgarcia  staff      24398 Oct 16 02:57 lodash-4.17.15.js.gz
-rw-r--r--  1 mgarcia  staff      13542 Oct 16 01:50 lodash-4.17.15.min.js
-rw-r--r--  1 mgarcia  staff  280028085 Oct 16 01:50 lodash-4.17.15.weird.js
-rw-r--r--  1 mgarcia  staff    3226049 Oct 16 01:51 lodash-4.17.15.weird.js.gz
-rw-r--r--@ 1 mgarcia  staff       7114 Oct 16 01:44 modernizr-custom.js
-rw-r--r--  1 mgarcia  staff       2689 Oct 16 02:57 modernizr-custom.js.gz
-rw-r--r--  1 mgarcia  staff       1296 Oct 16 01:45 modernizr-custom.min.js
-rw-r--r--  1 mgarcia  staff   14293251 Oct 16 01:45 modernizr-custom.weird.js
-rw-r--r--  1 mgarcia  staff     168443 Oct 16 01:45 modernizr-custom.weird.js.gz

Here's a table:

File	Source	Weird	Weird.gz	Minified	GZIP
Dommy	19648	51015234 (2596x)	587229 (29x)	8762 (0.44x)	5379 (0.27x)
Lodash	115023	280028085 (2434x)	3226049 (28x)	13542 (0.11x)	24398 (0.21x)
Modernizr	7114	14293251 (2009x)	168443 (23x)	1296 (0.18x)	2689 (0.37x)

So what? (part 2)

So, yeah - this isn't a good idea. If the Weird transpiler only changes the encoding of each character with a really weird equivalent, it makes a lot of sense that it doesn't compress better than the source one - the ideal scenario would be to compress the same.

I really didn't expect to get great results out of this, but on the other hand it was a nice opportunity to scratch some curiosity that I got - while pushing a bizarre, extreme Internet semi-joke a little bit more still. And no kittens were hurt during the making of it, so...