While not strictly necessary, replacing the two ASCII tab characters in the markdown copyright comment makes it easier to detect problematic TAB indentation within markdown files. After this commit, only 90 markdown files have leading ASCII TAB characters. And again, while not strictly necessary, we can consider running expand on those 90 markdown files to see what effect that might have on the HTML files they produce. Replace leading ASCII TAB characters with multiples of 8 ASCII SPACE characters (the default action of the `expand(1)` tool) might have an negative effect on some HTML formatting where **hack-a-round** formatting was attempted to fix broken indentation caused by problematic ASCII TAB characters. Think of this as a baseline for evaluating leading TAB characters within markdown files. TODO: Consider either using `expand(1)` to replace leading ASCII TAB characters with multiples of 8 ASCII SPACE characters AND/OR modifying the appropriate `bin` tool to convert leading ASCII TAB characters into multiples of 8 ASCII SPACE characters.
To build:
make
There is an alternate version that will read from the compiled file itself, in
case you don't have a /dev/urandom file. There is also a version that should
theoretically work with Windows which distinguishes binary and text. See
alternate code below.
To use:
./nyaruko [seed.txt] < original.bin > output.c
bash output.c > key.c
perl output.c > data.c
cat key.c data.c > output.c
cc output.c -o output
./output > regenerated.bin
NOTE: if your OS does not have /dev/urandom then you should specify a seed
file or use the alternate version which reads in the compiled
binary itself.
Try:
./try.sh
The script try.sh will check if you have perl before trying to use the perl parts of the script.
Alternate code:
The alternate version will, if no file is specified, read in the compiled binary
itself rather than trying to open /dev/urandom, which is the default for the
original entry. This is useful if your system does not have a /dev/urandom
file and you do not want to specify an extra file. The second alternate version
is like the first except it also should theoretically work with Windows, setting
binary mode on stdin and stdout.
Alternate build:
For the first version:
make alt
For the second version:
make alt2
Alternate use:
Use nyaruko.alt or nyaruko.alt2 as you would nyaruko above.
Alternate try:
For the first alternate version:
./try.alt.sh
We have no way to test the second version, sorry (tm Canada :-) )!
Judges' remarks:
The judges have nothing to add that has not already been written about in the spoiler.html (requires JavaScript) file! :-)
Author's remarks:
Usage
Nyaruko is a binary to text filter. Given some input on stdin,
Nyaruko will produce C code that reproduces this input on stdout:
./nyaruko < original.bin > output.c
gcc output.c -o output
./output > regenerated.bin
Output is encrypted, but both key and data are included in the output. To separate the key from the data, run these commands:
bash output.c > key.c
perl output.c > data.c
The key-less data.c still compiles, but produces a different message
on stdout instead of the original input. This message is a hint to
why the code is formatted the way it is.
To combine the key and data, concatenate them together in either order:
cat key.c data.c > output.c
cat data.c key.c > output.c
By default, Nyaruko generates a unique random key for every message,
using /dev/urandom as the seed. If given an extra command line
argument, Nyaruko will seed using that file instead of /dev/urandom:
./nyaruko seed.txt < input.bin > output.c
This makes the output key deterministic, allowing the same key to be shared
across different files. On operating systems that do not have /dev/urandom,
users should always specify this extra seed argument to avoid indeterministic
keys, or else use the alternate version which will by default read the compiled
alternate version itself, which might or might not be deterministic, depending
on where and how it's used.
Features
Implementation details that make Nyaruko more obfuscated than usual
programs:
Nyarukorecycles variables and buffers to reduce internal state. The variable names are also carefully chosen for mixed-case madness.Nyarukoemploys some preprocessor magic to share much of the same code and state between the encoder and decoder, and to increase occurrences of smileys ;)
Output code has these features:
- Fits nicely under 80 columns, and does not contain any trigraphs.
- Uses a fairly efficient encoding scheme, better than
uuencodefor files larger than ~13K, and better thanbase64for files larger than 20K. - Encryption with ISAAC, a cryptographically secure pseudorandom number generator.
- Trivial 3 language polyglot.
Code layout is meant to resemble
Nyaruko, also known as
Nyarlathotep, the Crawling Chaos. The most obvious thing to do with chaos is
to make a random number generator, and the most obvious thing to do with a
random number generator is to make
one-time-pads for encryption.
Compatibility
Nyaruko has these environment dependencies:
- Requires ASCII character set.
- Assumes
sizeof(unsigned int) == 4. - Best viewed with tab stops set to 8 spaces.
Output code has the same dependencies, with the additional requirement that the compiler must support arbitrarily long string literals. Maximum input size that can be encoded while still producing standard-compliant output is ~276 bytes for C89, and ~3168 bytes for C99.
Nyaruko has been to verified to work with these compiler/OS
combinations:
- gcc 4.4.5 on Linux (32bit and 64bit)
- clang 3.1 on Windows (Cygwin)
- gcc 4.5.3 on Windows (Cygwin)
- gcc 4.5.3 on Windows (MingW)
- gcc 4.3.5 on JSLinux
- tcc 0.9.25 on JSLinux
Note that on MingW, stdin and stdout are not opened in binary mode by
default, this means Nyaruko may not faithfully encode files on MingW.
Extra files
spoiler.html - Contains full recording of how the code went from blank state to an obfuscated program, gzipped to fit under 1MB. Requires JavaScript.