Dave Burton ioccc@snox.net
http://snox.net/ioccc
make
./prog "expression"
or
./prog
(input expressions)
./prog '?0x3343<<2'
echo '?0x3343<<2' | ./prog
./prog
Enter:
a=0xfeedface
b=0xcafe
a << 16 + b
grep '[0-9]' prog.c
./prog < test.in
less calc.man
“Look, Ma, no hands digits!”
The place where some arithmetic operations are done, for example, division or modulo, are easy to find. Where are addition and subtraction?
We were impressed with the level of detail the author “blogged” the obfuscation process.
All useful tools should include sufficient documentation. See the included man page. ;-)
This program is named calc. Onomatopoeia of sorts, in the visual realm. The formatting was easier: Form Follows Function (although Hou did this better). It was tempting to call it iCalc, with i for integer (and it was written on an iDevice), but sanity prevailed.
Printed in 8 pt Courier, it makes aesthetically pleasing 8.5 x 11 wall art.
Dominik Muth observed a syntax error on 1+1
, 1-1
, 1|1
, and 1^1
when using an
ARM based computer, although 1*1
and others worked correctly.
This occurs because some platforms by default treat char
as unsigned char
.
The original code has this weakness, which can be corrected by adding -fsigned-char
to the compile line.
The modified code is agnostic to char
signed-ness.
Yes, this is another calculator program.
However, unlike any other calculator source code the author is aware of, this one contains no digits. At all. Anywhere. As in
grep '[0-9]' prog.c || echo no digits!
Nor are the digits simply obfuscated. There are no character constants. There is only one single, short string, and it does not contain any obfuscated digits. And yet the output is in decimal and hexadecimal, and it accepts octal, decimal, and hex input.
Flow control is by for
and return
. It is If-less. Switch-less. While-less. Hope-less.
Literally, if not figuratively. ;-)
The code size is intentionally 2015, as stated by iocccsize -i, and can be reformatted without changes (as in rule2.c) to exactly 4096 octets.
Finally, calc is interesting, if for no other reason than it is practically useful: Hex/decimal conversion, arithmetic, byte-swapping, bit-shifting, and logical operations with memory, on the command line, or interactively.
The code follows many industry best-practices:
You know what it does just by looking at it.
Numbers and digits have often been used for obfuscation, so in a bid for better clarity, this code has no digits at all.
In the spirit of clarity, most of the variables are declared before use, so they are readily visible, with single letter names that don’t clutter the screen, in a tautological arrangement.
More helpful is the use of typedef
s instead of #define
, so the types remain useful
even in the debugger, and source code beautifiers will not undo careful presentation.
#defines
are used here in the best spirit of the preprocessor: to shorten
repetitive constructs, not for obscurity.
All functions are short (and therefore easy to understand?) and do only one thing, well, ….
Curly braces invite endless holy wars over readability. This code only uses them for function definitions, thereby greatly improving readability, and reducing arguments about said readability.
One of C’s many blessings (until recently) is a modicum of keywords,
of which only 8 are used:
for
, return
, typedef
, void
, char
, short
, unsigned
and long
;
10 if #include
and #define
are counted.
The code helpfully declares “syntax_error!” right up front, filling the reader with confidence.
In spite of its inherent clarity, the code was presented to both indent(1) and clang-format(1), after running it through the preprocessor, sustaining its better qualities.
That the code is obfuscated is spelled out two different ways in the source; it helps to reformat the code to see the second spelling.
The author is overly fond of the code that handles input. There is no memory allocation, nor pre-allocation of arrays, so where does the memory come from? The multi-base strtol() code is pretty trick. No special attempt was made to replace operations with obscure mathematical relations, as the (lack of) beauty is in the parsing – ask not where is the modulus operation, but how does it know when to mod – okay, but where is add and subtract? And really, who doesn’t enjoy a setjmp() in the argument list?
Finally, documentation of the source code structure is helpfully provided in rot13 format. But isn’t it unsportsmanlike to rely upon it? However, even with this helpful guide, things like reading input lines remain as tricky to read as they are to write.
Calc is C99 compliant, and is ANSI/C90 compliant except for wanting 64-bit long long
.
Changing the two long long
typedefs to long
, and compiling with gcc -m32
yields a program that operates on 32-bit values, with wrap-around of shifts mod 32:
1<<48
65536 0x00010000
gcc 4.2.1 without arguments compiles cleanly (e.g. cc prog.c -o prog). This is my workaday compiler. gcc is also clean with -ansi, and -std=c99.
clang 3.6.0 lives up to its namesake, resounding on non-problems. To stifle these complaints: -Wno-return-type -Wno-parentheses -Wno-empty-body
When invoked with -ansi -pedantic, there are two warnings, which can be ignored:
ISO C90 does not support 'long long'
With -Wall -std=c99, both are quite noisy, and all can be ignored:
array subscript has type 'char' [-Wchar-subscripts]
value computed is not used [-Wunused-value]
operation on 'j' may be undefined [-Wsequence-point]
If the program name begins with an “e”, it echoes stdin to stdout. This allows for a convenient test suite:
ln -s prog eprog
PATH=. eprog < test.in | diff - test.out
If you do not want to puzzle out how it works, see spoilers.markdown.
© Copyright 1984-2016,
Leo Broukhis, Simon Cooper, Landon Curt Noll
- All rights reserved |