blog · git · desktop · images · contact & privacy · gopher

---

Moving the first line to the end of the file

2017-07-03

Consider a very simple file:

foo 1
bar 2
baz 3
yadda 4
whatever 5

You know want to edit this file in-place, moving the first line to the end of the file. The expected result is this:

bar 2
baz 3
yadda 4
whatever 5
foo 1

How do you do this in a shell script?

My first idea was using sed:

sed -ni '1{h;n}; $!p; ${p;x;p}' data

It goes like this:

• If you’re on the first line, copy pattern space to hold space and then move on to the next line (i.e., print nothing).
• If you're not on the first line and not on the last line, print it.
• If it’s the last line, print it, fetch hold space, and then print that.

It works, but it’s not so easy to understand. You have to know about the hold space and what not. I dropped this solution, because I didn’t want to “impose” it on my colleagues.

Alright, let’s try a very explicit approach. How about this one:

lines=$(wc -l <data)
if (( lines > 1 ))
then
    new_tail=$(head -n 1 data)
    all_but_first=$(tail -n +2 data)

    cat >data <<EOF
$all_but_first
$new_tail
EOF
fi

That’s easy to understand, isn’t it? It also shows a flaw in my sed command above: It fails if there is only one line in data. Hence the if.

Since I still had the discussion about bad taste in mind, I wasn’t happy with this solution, either. That if shouldn’t be needed.

Isn’t there something nicer?

Let’s try awk.

awk 'NR == 1 { s = $0 }; NR > 1 { print }; END { print s }' data

That’s better. You can make it shorter by removing the { print } block, because that’s the default action.

Problem is, awk cannot edit files in place. So, the actual call looks like this:

tmpfile=$(mktemp)
awk '...' <data >"$tmpfile"
mv "$tmpfile" data

Ugh, that’s pretty verbose. Plus, if the script was to fail for whatever reason at the awk step, there’d be a dangling temporary file. To avoid this, we would have to install a trap …

Still not happy.

It bothered me that this task is so easy to do in vi, so why has it to be hard in a shell script? Wait a second. Why not use ed?

ed -s data <<END
1m$
wq
END

That’s not bad. It has the same problem as sed: Very unusual commands. Of course, they are related. But there is no corner case. And it’s just one command, really. Pretty much anyone will understand wq.

I like that.

Comments?