I am on GNU bash, version 4.3.11.
Say I want to print unique lines on a file. I am using this approach, which works well on a file:
$ cat a
9
10
9
11
$ awk '!seen[$0]++' a
9
10
11
However, if I get the input from stdin, using double quotes in a multi-line and piping to awk, it fails:
$ echo "9
> 10
> 9
> 11" | awk '!seen[$0]++'
bash: !seen[$0]++': event not found
That is, bash tries to expand the command seen, which of course does not know because it is a variable name. But it shouldn't happen, since the command is placed within single-quotes.
echoing in a single-quoted, multi-line input works well:
$ echo '9
> 10
> 9
> 11' | awk '!seen[$0]++'
9
10
11
The funny thing is that it also works well on a single-line input that is double-quoted:
$ printf "9\n10\n9\n11" | awk '!seen[$0]++'
9
10
11
I wonder why is Bash trying to expand history if it occurs after a mutiline input, even though the command itself uses single-quotes.
Other considerations:
Having a pipe in between does not fix it, either:
$ echo "9
> 10
> 9
> 11" | cat - | awk '!seen[$0]++'
bash: !seen[$0]++': event not found
And setting set +H turns history off, so it works well because it does not try to expand anything:
$ set +H
$ echo "9
> 10
> 9
> 11" | awk '!seen[$0]++'
9
10
11
I went through the canonical answer by rici on how to address error “bash: !d': event not found” in Bash command substitution and found many possible reasons, but none matches this behaviour.
It is not a bug.
After asking this in the bash-bugs mailing list, I got the following answer:
History expansion is explicitly line-oriented.
It doesn't know about shell state, especially shell quoting state, that spans lines.
It does know about vaguely shell-like quoting that's common across a large set of Unix utilities -- since the history and readline libraries are used outside the shell -- and that a double quote introduces a quoted string in which single quotes are not significant and don't inhibit history expansion.