A quine in sed

2019-04-01 / 2 mins read

Yesterday someone on IRC corrected their typos with classic sed-like replaces¹, and for some reason that got me wondering if you can write a quine in sed.

Of course you can.

echo "s/\\\\/&&&&/g;s/\"/\\\\&/g;s/.*/echo \"&\" | sed \"&\"/" | sed "s/\\\\/&&&&/g;s/\"/\\\\&/g;s/.*/echo \"&\" | sed \"&\"/"

When run in dash or zsh (but not bash²), the output is that same line. That's what quines are: self-replicating programs. For a thorough review, see David Madore's excellent page about quines.

Dissection

Like many quines, this one consists of a data part and a stream manipulation part. That's necessary to use sed in the first place. However, the escaping mechanism of the shell also plays a large part, so strictly speaking this is a POSIX shell quine that uses echo and sed.

The data part is what is echoed and piped to sed. The sed command run on the input is this:

sed "s/\\\\/&&&&/g;s/\"/\\\\&/g;s/.*/echo \"&\" | sed \"&\"/"

This combines three regex replace commands:

1. s/\\\\/&&&&/g
2. s/\"/\\\\&/g
3. s/.*/echo \"&\" | sed \"&\"/

The third one is really the only interesting part: take the input, and basically print echo <input> | sed <input>. Figuring out that part is easy; finding a way to work with character escaping such that the input data can be crafted properly is a bit more time-consuming.

I tried finding a smart way of combining ', ", \x22 and such, but that turned out to be difficult. Working with the shell escaping mechanism instead of avoiding it was a better idea. Just find a way to escape special characters before printing. Then write the program as data. Voila, the quine is ready.

$ echo "s/\\\\/&&&&/g;s/\"/\\\\&/g;s/.*/echo \"&\" | sed \"&\"/" | sed "s/\\\\/&&&&/g;s/\"/\\\\&/g;s/.*/echo \"&\" | sed \"&\"/"
echo "s/\\\\/&&&&/g;s/\"/\\\\&/g;s/.*/echo \"&\" | sed \"&\"/" | sed "s/\\\\/&&&&/g;s/\"/\\\\&/g;s/.*/echo \"&\" | sed \"&\"/"