What encoding Facebook uses in JSON files from data export?

Question

I've used the Facebook feature to download all my data. The resulting zip file contains meta information in JSON files. The problem is that unicode characters in strings in these JSON files are escaped in a weird way.

Here's an example of such a string:

"nejni\u00c5\u00be\u00c5\u00a1\u00c3\u00ad bod: 0 mnm Ben\u00c3\u00a1tky\n"

When I try parse the string for example with javascript's JSON.parse() and print it out I get:

"nejnižší bod: 0 mnm Benátky\n"

While it should be

"nejnižší bod: 0 mnm Benátky\n"

I can see that \u00c5\u00be should somehow correspond to ž but I can't figure out the general pattern.

I've been able to figure out these characters so far:

'\u00c2\u00b0' : '°',
'\u00c3\u0081' : 'Á',
'\u00c3\u00a1' : 'á',
'\u00c3\u0089' : 'É',
'\u00c3\u00a9' : 'é',
'\u00c3\u00ad' : 'í',
'\u00c3\u00ba' : 'ú',
'\u00c3\u00bd' : 'ý',
'\u00c4\u008c' : 'Č',
'\u00c4\u008d' : 'č',
'\u00c4\u008f' : 'ď',
'\u00c4\u009b' : 'ě',
'\u00c5\u0098' : 'Ř',
'\u00c5\u0099' : 'ř',
'\u00c5\u00a0' : 'Š',
'\u00c5\u00a1' : 'š',
'\u00c5\u00af' : 'ů',
'\u00c5\u00be' : 'ž',

So what is this weird encoding? Is there any known tool that can correctly decode it?

Answer 1

Thanks to Jen's excellent question and Shawn's comment.

Basically facebook seems to take each individual byte of the unicode string representation, then exporting to JSON as if these bytes are individual Unicode code points.

What we need to do is take last two characters of each sextet (e.g. c3 from \u00c3), concatenate them together and read as a Unicode string.

This is how I do it in Ruby (see gist):

require 'json'
require 'uri'

bytes_re = /((?:\\\\)+|[^\\])(?:\\u[0-9a-f]{4})+/

txt = File.read('export.json').gsub(bytes_re) do |bad_unicode|
  $1 + eval(%Q{"#{bad_unicode[$1.size..-1].gsub('\u00', '\x')}"}).to_json[1...-1]
end

good_data = JSON.load(txt)

With bytes_re we catch all sequences of bad Unicode characters.

Then for each sequence replace '\u00' with '\x' (e.g. \xc3), put quotes around it " and use Ruby's built-in string parsing so that the \xc3\xbe... strings are converted to actual bytes, that will later remain as Unicode characters in the JSON or properly quoted by the #to_json method.

The [1...-1] is to remove quotes inserted by #to_json

I wanted to explain the code because question is not ruby specific and reader may use another language.

I guess somebody can do it with a sufficiently ugly sed command..

Answer 2

The encoding is valid UTF-8. The problem is, JavaScript doesn't use UTF-8, it uses UTF-16. So you have to convert from the valid UTF-8, to JavaScript UTF-16:

function decode(s) {
   let d = new TextDecoder;
   let a = s.split('').map(r => r.charCodeAt());
   return d.decode(new Uint8Array(a));
}

let s = "nejni\u00c5\u00be\u00c5\u00a1\u00c3\u00ad bod: 0 mnm Ben\u00c3\u00a1tky\n";
s = decode(s);
console.log(s);

https://developer.mozilla.org/docs/Web/API/TextDecoder

Answer 3

Just adding the general rule how to get from something like '\u00c5\u0098' to '?'. Putting together the last two letters from the \u parts gets you c5 and 98 which are the two bytes of the utf-8 representation. UTF-8 encodes the code point in two bytes like this: 110xxxxx 10xxxxxx, where x are the actual bits of the character code. You can take the two bytes, use & to get the x parts, put them one after the next and read that as a number and you get the 0x158, which is the code for '?'.

My javascript implementation:

    function fixEncoding(s) {
        var reg = /\\u00([a-f0-9]{2})\\u00([a-f0-9]{2})/gi;
        return s.replace(reg, function(a, m1, m2){
            b1 = parseInt(m1,16);
            b2 = parseInt(m2,16);
            var maskedb1 = b1 & 0x1F;
            var maskedb2 = b2 & 0x3F;
            var result = (maskedb1 << 6) | maskedb2;
            return String.fromCharCode(result);
        })
    }

Answer 4

You can use a regular expression to find groups of almost unicode characters, decode them into Latin-1 and then encode back into UTF-8

The following code should work in python3.x:

import re

re.sub(r'[\xc2-\xf4][\x80-\xbf]+',lambda m: m.group(0).encode('latin1').decode('utf8'), s)