Julia Evans

Day 11: How does gzip work?

Spoiler: I don’t really know yet, but there is a lot of mucking with bits.

Yesterday I was flailing around a bit looking for projects. Today at lunch Kat helped me figure one out: writing a parallel version of gzip in Julia. Julia is a pretty good choice for this because it lets you do low-level bit-fiddling and write efficient algorithms, but still has lots of nice high-level features. It also looks a lot like Python!

The document I’m using to understand how gzip works is this really detailed and wonderful page. So if you actually want to know you should just read that.

The basic idea behind gzip is

  1. use some not-so-complicated algorithms (Huffman coding, LZ77 compression). These things sound like they would be the complicated part, but so far they don’t seem to be too bad. Conceptually. I think.
  2. Do all kinds of bit fiddly things to actually make it efficient.

Some choice things from the article (emphasis mine)

  • “hclen is the declaration of four less than how many 3-bit length codes follow” (why four?!??)
  • Everything is variable-length encoded, so instead of just having bytes, you have bit sequences of various lengths that you have to extract. Because efficiency. Huffman coding is what makes this variable-length encoding madness actually work.
  • Every gzip file begins with the “magic bytes” 1F8B

Here is a snippet of Julia code that I wrote to work on this! This code takes an array of 8 bits and converts it into a byte. By default Julia does some smart things like bounds checking, but you can make this even faster by preventing bounds checking with @inbounds.

You can see that this kind of just looks like Python, except it is fast! (I promise)

function make_int(bv::BitVector)
    num = 0x00
    for i=1:length(bv)
        num = (num << 1) + bv[i]
    return num

That’s it! Maybe tomorrow I will actually understand how gzip uses Huffman coding. So far I have started to decode the gzip header.