Entries
Winning entries Winning authors Location of authors Bugs and (mis)features Fixing entries Updating author info Thanks for the help
Status
News Contest status Rules and guidelines Markdown guidelines Security policy
FAQ
Frequently Asked Questions How to enter Compiling entries Running entries How to help
About
Home page About the IOCCC The Judges Contact us
IOCCC image by Matt Zucker

The International Obfuscated C Code Contest

2018/burton2 - Best abuse of the rules

Tokenize and count

← 2018/burton1 ↑ 2018 ↑ 2018/ciura → C code Makefile Inventory ✓

Author:

To build:

    make

To use:

    ./prog [-tcksri] < file.c

Try:

    ./try.sh

If you get really stuck, try:

    man ./tac.1

Judges’ remarks:

They say size isn’t everything, and in the case of IOCCC iocccsize.c that is saying something! What is this program weighing and how much does it weigh?

Author’s remarks:

tac - tokenize and count C and derivative languages:

tac computes C program size
(obfuscated / otherwise)
by splitting code as tokens small,
ignoring space, then counting all.

It tries quite hard to act the same
as counting tool of I-OCCC fame.
Some bugs were fixed, it’s faster too,
and does more things, just add some glue!

It does far more than print the size:
the token part’s a better prize!
With each upon on a single line,
how many things can you design?

Included is a keyword sorter,
de-obfuscator, freq reporter.
With code produced as little parts,
just add your own creative arts.

But first you have to figure out
the braces part, without a doubt!
And then the code - it isn’t easy:
for as else? That’s rather cheesy!

The rest of it is standard fare:
expressions strange, the globals bare,
confusing symbols one and oh,
and precedence you have to know.

The keyword list was made external,
because of standards change eternal.
Thus other languages can now,
be counted just like C, somehow!

I hope you find this code obscure
enough to win, and thus procure
a public place for all to see
how badly I can butcher C!

Said another way….

tac is the inverse of cat(1): it un-concatenates its input into C tokens, writing the token stream to stdout. There are options to suppress the token ids, summarize the token counts, and several options to allow it to be compatible with iocccsize.

The token-based output is quite useful in analyzing source code (see below), and the counter is more accurate than iocccsize (see below); but this can be fixed by using the -c compatibility flag, restoring the following problems as in the official tool:

The following bugs are always corrected:

There is one (known) remaining problem with detection:

    #/*this will not be counted*/include/*correctly*/<stdio.h>

Fixing this properly hampers backwards compatibility in counting. This requires eliding comments during the translation phases (ISO 5.1.1.2, phase 3), and yet still counting words correctly absent -k.

Backwards Compatibility:

tac was run over all 366 previous IOCCC winning entries:

    find ~/src/obc -type f -name \*.c | wc

The discrepancies found are documented and explained in the file discrepancies.html.

    find ~/src/obc -type f -a -name "*.c" | xargs ./spotcheck.sh ./prog | ./spotdiff.sh |
            grep -v keep | diff -bw - discrep* | grep "[<>] cl "

In summary, there are only 6 unique entries out of 366 that have any variation in Rule 2 counts; in all cases, the differences are due to bugs within iocccsize, described in detail in the file discrepancies.html. There are only 40 files (40/366, ~11%, including originals and variations) that differ at all, 31 only in word count; most word count differences appear related to comment block detection, weighted more towards recent entries (2011 and onward = 20/40 ~50%).

This (minor) variation in word count values has not been addressed since it is not a qualifying metric, nor is it very useful in a programming context: whitespace surrounds English words, not necessarily C tokens, as in a+=1;.

Accuracy (versus compatibility) is demonstrated through a series of small test cases designed to allow exhaustive, hand-counted values for comparison with the tool values. These test cases were vital in debugging and regression testing, and provided a way to determine which tool was correct when there were differences.

A version of this tool in more clearly written C
is presented for the Judge’s consideration as a more accurate replacement for iocccsize.
This obfuscated entry is derived from (and compatible with) that code,
but due to obfuscation, has had some significant, deep changes for the contest.
Nevertheless, the more clearly written code remains a spoiler for this entry.

NB: iocccsize gets a different answer from tac on its own iocccsize.c source code; tac gets the correct answer. This is due to the aforementioned bugs within iocccsize, proved by fixing iocccsize.c with the included patch, so iocccsize reports the correct answer for itself.

But wait… There’s More!

There is no limit on line length, file length, comment length, or identifier length.

tac features an accurate tokenizer; which presents each C token on a line by itself. This feature enables interesting analyses of C (and C++) code that’s hard to get any other way.

For instance, here is a simple token counter, useful in finding repeated long tokens, or operator frequency, or counting the references to identifiers, constants, or breadth of use of the language:

    #!/usr/bin/env bash
    cat $* | ./prog -t | sort | uniq -c | sort -k1nr

And here is a C keyword frequency counter:

    #!/usr/bin/env bash
    function iskeyword {
       awk 'BEGIN{f="c11";while(getline<f)k[$1]=0} {if($1 in k)k[$1]++}END{for(i in k)print k[i],i}'
    }
    cat $* | ./prog -t | iskeyword | sort -k1nr

This script was used to “optimize” the reserved word order so the most frequent IOCCC winning entry keywords are checked first.

An interesting aside: as might be expected, obfuscation has changed the frequency of keyword distribution over time. The top five for the decades:

    1980s       1990s       2000s       2010s
    -----       -----       -----       -----
    94 if       532 char    375 if      254 int
    69 for      524 if      372 char    217 for
    59 while    417 int     332 int     198 if
    48 char     223 for     184 for     121 return
    43 int      223 void    165 return  104 char

The keyword list is external to the program, and is easily changed, sorted, checked, verified. This allows such additional programs in concert with the tool’s primary operation.

A poor-man’s de-obfuscator can be based upon the output of tac -t and a handful of simple rules (and another use of the external keyword file). A more refined version of this is included in the file unob.sh, but the simple code below is a serviceable obfuscated C de-obfuscator in a scripting language. It really is this easy with tac:

    #!/usr/bin/env bash
    script='
    BEGIN {
            last=nl="\n";
            f="c11"; while(getline <f > 0) kw[$0]++; close(f);
    }

    function iskw(a)   { return a in kw }
    function indent(a) { return sprintf("%*s", n*3, " ") }
    function newline() { if (!infor && last != nl) printf last=nl; }
    function show(a) {
            if (last==nl) printf "%s", indent()
            printf "%s%s", space(), a
            last=a
    }
    function space() {
            return iskw(last) ||
              (last ~ /[A-Za-z0-9_+-\/%^[&\]\)=:<>;]$/ && $0 !~ /[:;()\[\],]/) ? " " : ""
    }

    /^\(/       { ++paren }
    /^\)/       { --paren }

    /^for/                  { newline(); infor=1 }
    infor && /^;/           { ++infor; show($0 " "); next }
    infor==1 && /^:/    { ++infor }
    infor>1 && paren==0 { infor=0 }

    /^\?/               { tern++; n++; newline(); show($0 " "); next }
    tern && /^:/        { newline(); show($0); --tern; --n; next }

    /^#/        { newline(); show($0); newline(); next }
    /^;/        { show($0); newline(); next }
    /^{/        { show($0); ++n; newline(); next }
    /^}/        { n--; newline(); show($0); newline(); next }

    { show($0) }
    '
    cat $* | sed 's/#include/##include/' | cpp -E -trigraphs |
    sed 's/^# .*$//' | sed 's/^#//' |
    ./prog -t | awk "$script"

On the question of obfuscation:

Rule 2:

The keyword list is externalized as an include file. This presents a useful feature for the source code: not only can the reserved word script above be produced, but the tool can be rebuilt with any set of reserved words, just by using a different list. For instance: removing all the secondary keywords; removing all the C11 keywords; trying just K&R C. Since C++ and Java share the same operators as C, just change the keyword list and tac will correctly tokenize both of these; the tokens >>> and :: can be handled with a short post-filter (tokenfix.sh, included).

I believe this entry may also satisfy the request for gratuitous use of all the C11 keywords? Whether this context satisfies “intended C language context” is a matter of interpretation. ;-) And tac is ready for the IOCCC++ contest. And the IOJCC trash can.

The following reserved word files are included:

NB: The keyword file used in this code is derived from the list in iocccsize.c, which is neither complete (#define, #ifndef, #undef are missing

    yes, Virginia knows about `#define` omitted on purpose

nor correct (many more are added: I, true, bool, compl, …):

    (sed -n '1p' c11; sed -n '2,$p' ioccc.kw.freq | sort) | comm -3 - c11

That is an obfuscated way to say “diff”. But it also more clearly shows the, um, diffs.

The usage string is similarly externalized, providing a better usability string than the limits would otherwise allow. If this is considered bending the rules too far, just -DU=O to get a passible, if cryptic, usage message using no additional octets than without the define. This also allows the builder to configure the amount of help a utility should provide (you want a manpage? Because that’s how you get a manpage.) – a rather significant point of contention made moot.

There is definitely abuse of the Rule2 counter, exploiting a “feature” of iocccsize: iocccsize does not Rule2 count {, } and ; when followed by whitespace – even within a string! Thus: Freecode, a simple, uncounted, highly obfuscated encoding of a central part of this program. (The reserved word list is not Freecoded, both because it is useful apart from the program, and because Freecode is not entirely free.)

Compilation:

The code is ANSI compliant, but it does have a string longer than 509 characters. It has been tested on OSX and Linux, Clang and GCC.

The code uses trigraphs, ironically: it parses trigraphs with trigraph code. Because that’s funny. And it remains obscure after, so trigraphs are not used for obfuscation. Thus -trigraphs -Wno-trigraphs (which is also humorous).

It’s obfuscated C. Thus -Wno-parentheses -Wno-empty-body -Wno-char-subscripts.

The code uses unsigned chars, but C strings are signed. Thus -Wno-pointer-sign. NB: This is not a portability concern, the one string holds only printable ASCII characters.

Thus:

    cc -ansi -Wall -trigraphs -Wno-trigraphs -Wno-parentheses -Wno-empty-body -Wno-char-subscripts -Wno-pointer-sign -DU=O -DW=\"keywords\" -o prog prog.c

Final remarks:

Cody Boone Ferguson was relentless in his pursuit of bugs. Thanks to his reports, the version of unob.sh is stronger, tac groks digraphs, tokenfix.sh corrects for missing digraphs in prog.c, and the program manpage.sh was added to the corpus.

manpage.sh is a useful program that turns ASCII versions of man pages (e.g., tac(1) man page) into real man pages. It works for C and C-like languages, command lines, and was designed to handle man page sections 1-8. It works well on most inputs, but tends to need some raw nroff for more fancy constructions, such as combined arguments (e.g. the -r and -s arguments to tac are not recognised when written as -rs in prose).

Try:

    ./manpage.sh tac

    ./manpage.sh -h
    ./manpage.sh -h | ./manpage.sh
    ./manpage.sh -h | ./manpage.sh -R | less
    ./manpage.sh -h | ./manpage.sh -P | lpr

NOTE: in 2023 the tac.man was updated to proper man format and renamed tac.1.

Inventory for 2018/burton2

Primary files

Secondary files

Jump to: top

Coda

Nu HTML check this web page
Jump to top Jump to Content Jump to Inventory