Best one-liner

Judges' comments:

To use:

make
./prog < any-file

Try:

./prog < prog
./prog < prog.c

Selected Judges Remarks:

One line, one array, one loop and one statement but it prints many bytes. It won.

Author’s comments:

Feed this anything on stdin.

The included script may be used as a driver program for this entry. It adds several nice-to-have features, and allows salt-to-taste customizability.

This code has been compiled and tested on:

OSX-10.11, gcc 4.2.1 and clang 3.0, -O0 and -O3
X86-Linux, gcc 4.1.2 and 4.8.4, -O0 and -O3
ARM-QEMU, gcc 4.8.4, -O0 and -O3
X86-FreeBSD 10.3, clang 3.4.1, -O0 and -O3
TCC 0.9.26 (x86-64) (used as a proxy for 2001/bellard)
UNIX V7 pcc (Johnson’s C compiler) on a simulated PDP-11/45 (see spoiler for further detail)

Spoiler below; scroll down.

SPOILER

This started life as a personal challenge: write the shortest hex dump utility in C.

The self-imposed ground rules in the hunt for shortest: the code has to output correctly, no extraneous newlines, the code must be portable, and without bypassing the solution using execve(2), or functions built upon it (system(3), popen(3), etc). E.g:

main(){system("hd");}   // it's turtles all the way down!

is not acceptable. And, “output correctly” is the output format of this entry. QED.

Early attempts ended up at a reasonably short 157 characters. This was too long for a one-liner, so it had to be formatted as something more interesting. This solution is pleasing, but a true one-liner was the hoped-for goal. Here is the 157 character version in it’s entirety:

char     O,o[  20];L(O){
putchar  ( O+  48+39  *(9<
O));}    I(O)  {L(~     15);
O|| puts(o);}  main     (l){
for(;~(l=getchar())     ;I(O
&=15     ))o[  O++      ]=4<
(4^l     >>5)  ?l:     46,L
(l>>     4),L  (l&15);O&&
I(o[     O]=0  );}//157c

While thinking about this in the shower (you are not the only one that does this!) came the idea which reduced the code to just 119 chars. This dramatic reduction started to feel close to minimal, and the hunt resumed. The progression to the present size is interesting, and it is interesting that the same algorithm can be expressed so differently by slight rearrangement of its parts.

// 119
char O,o[20];main(l){for(;~(l=getchar());O||puts(o))o[O++]=isprint(l)?l:46,O&=15,printf("%02x ",l);O[o]=0;O&&puts(o);}
// 116
char O,o[20];main(l){for(;O++[o]=~(l=getchar())?4<(4^l>>5)?l:46:0;(O&=15)||puts(o))printf("%02x ",l);O>1&&puts(o);}
// 115
char O,o[20];main(l){for(;O[o]=~(l=getchar())?4<(4^l>>5)?l:46:0;(O=-~O&15)||puts(o))printf("%02x ",l);O&&puts(o);}
// 113
char O,o[20];main(l){for(;~l;(O=~l?printf("%02x ",l),-~O&15:!O)||puts(o))O[o]=~(l=getchar())?4<(4^l>>5)?l:46:0;}
// 112
char O,o[20];main(l){for(;~l;(O=~l?-~O&printf("%02x ",l)*5:!O)||puts(o))O[o]=~(l=getchar())?4<(4^l>>5)?l:46:0;}
// 111
char O,o[20];main(l){for(;~l;O||puts(o))O=(O[o]=~(l=getchar())?4<(4^l>>5)?l:46:0)?-~O&printf("%02x ",l)*5:!O;}
// 110
char O,o[0];main(l){for(;~l;O||puts(o))O=(O[o]=~(l=getchar())?4<(4^l>>5)?l:46:0)?-~O&printf("%02x ",l)*5:!O;}
// 109
char O,o[];main(l){for(;~l;O||puts(o))O=(O[o]=~(l=getchar())?4<(4^l>>5)?l:46:0)?-~O&printf("%02x ",l)*5:!O;}

There are solutions without the ASCII sidebar that are significantly shorter. These do not compete in the same category, since their output is materially different. They are derived from the solutions above, but are still fascinating as concise utilities, so here is their progression, for the curious:

// 102
I;O(O){putchar(O+48+39*(9<O));}main(l){for(;~(l=getchar());O(++I&15?~15:~37))O(l>>4),O(l&15);O(-38);}
// 80
O;main(l){while(~(l=getchar()))printf("%02x%c",l,++O&15?32:10);O&15&&puts("");}
// 78
O;main(l){for(;~l;(O=~(l=getchar())?printf("%02x ",l),-~O&15:!O)||puts(""));}
// 77
O;main(l){for(;~l;(O=~(l=getchar())?-~O&printf("%02x ",l)*5:!O)||puts(""));}
// 76
O;main(l){for(;~l;O||puts(&O))O=~(l=getchar())?-~O&printf("%02x ",l)*5:!O;}

It is possible to reduce the size of each of these by one by eliding the trailing newline. This was not reported above since the newline kept getting appended by processing software, and it is more interesting to remove characters by changing code than by filtering:

$ tr -d '\012' < 109.c | wc -c
108
$ tr -d '\012' < 76.c | wc -c
75
$ tr -d '\012' < 109.c | cc -o 108 -xc -
$ verify 108
$

An obfuscated hex dump has been done before: 1986/bright/bright.c is similar in function, but it uses more characters in just setting up its obfuscation than this entire solution!

BIGGER SPOILER

If you are unhappy with the way DEL (character 127) is handled, replace the cryptic expression 4<(4^l>>5)?l:46 with 32>l|l>126?46:l. Both take the same number of characters, but the former is a bit more interesting. Bonus question: how does the first expression work?

BIGGEST SPOILER (and some history)

Technically, char o[0]; is illegal (SS 6.7.6.2) according to ISO (N1570). And char o[]; is a conditionally supported feature (SS 6.7.6.2, 6.10.8.3). However, it works in modern C implementations as an extension, including GCC, Clang and TCC, for X86 and ARM, on multiple operating systems (Linux, OSX, FreeBSD), so it is functionally correct. MicroSoft Visual C++ does not allow char o[] or char o[0], and implements its close relative char o[1] such that writing more than a single char will cause a fault, which is technically correct.

But how does this work at all, even as an extension? o[] allocates one(!) byte in a zero-filled data area (formerly BSS). The UNIX-derived systems this code was run upon all place the storage of O and o into a zero-filled VM page with nothing following: thus, there are plenty of available and unused zero bytes following O, addressed via o, and this code only needs 17. This is undefined behavior, but as shown, it is widely available.

The 1979 Seventh Edition UNIX (v7) system pre-dates the IOCCC by several years, and the source code to sh(1) was a significant inspiration for this contest. There are two C compilers present: cc written by Dennis Ritchie, and pcc written by Steven Johnson. The nested conditional expressions in 116 and smaller are too complex for cc (“Illegal conditional”) but pcc handles them correctly, and neither compiler accepts char o[] or char o[0] nor puts the bytes in a zero-filled page (as a paged VM was not yet written for UNIX).

More significant is that v7 printf(3) does not report the number of characters written, and therefore 111 compiles but does not work correctly; 113 is the shortest code that compiles and runs correctly in Ancient UNIX.

Interesting digression: v7 printf (_doprnt) was written in PDP-11 assembler. It pre-dates varargs.h, and assembly was the only way to gain access to the variable argument types. The first edition The C Programming Language, 1978, page 71, remarks:

By the way, there is no entirely satisfactory way to write a portable function that accepts a variable number of arguments, because there is no portable way for the called function to determine how many arguments where actually passed to it in a given call. … printf, the most common C function with a variable number of arguments, … is also non-portable and must be modified for different environments.

It is fascinating to explore the roots of C through the history of the operating system it was birthed within. C is often criticized as a difficult, inscrutable language, but as this contest demonstrates, this is in parody. Shell-gol is one of those instances where someone preferred a different language, and (ab)used the preprocessor to re-mold C into something they found comfortable. Most of the rest of the system, especially in the kernel, are models of clarity and efficient expression.

TL;DR

So which version is the shortest, portable hex dump?

109 works with (most?) modern Unix-like systems and compilers, see example list above.
111 works adds Microsoft, is standards-approved, and does not exploit undefined behavior.
113 works on Ancient UNIX and the portable C compiler.
119 works on all platforms, including the original Ritchie PDP-11 C compiler.

Coda

Clearly, the smallest possible program violates modern best practice. It is not even possible to compile cleanly, but it will compile correctly.

Included is prog.nowarn.c, a modern C implementation that is – perhaps – the smallest, clean-compiling version using clang or gcc with -Wall -pedantic. Note that while this is larger than all but the original attempt, it will not work with the original Ritchie PDP-11 C compiler, since the expression within the ternary remains too complex.

Try

cat prog.c prog.nowarn.c

to compare the differences (diff is not helpful on a one-liner).