How can I replace a literal \n with a newline character? (on OS X)

Question

I have a script which pulls something out of my shell history and saves it to a file, but if there is a newline in the history entry it gets returned to me with a literal \n instead of the newline character. I'd like to replace the \n with a newline before saving it to the file.

I'm on OS X, and figured the problem might be the BSD-based sed that ships with OS X (that always seems to be the problem when sed isn't behaving!) so I tried with gsed, and what was failing with sed worked on the first try with gsed:

history -1 | awk '{$1=""; print}' | gsed -e 's/^[[:space:]]*//' -e s/'\\n'/"\n"/

But if at all possible I'd like something that'll work with the default utilities (whether sed or something else) on both Linux or OS X, since I can't count on gsed being installed on OS X, and since GNU sed is the normal sed for Linux there is obviously no need for gsed.

Answer 1

sed 's/\\\\/& /g;s/\\n/\
/g;  s/\\\\ /\\\\/g' <in >out

...should handle the \newline replacements without misinterpreting backslash escapes - so you can still have a literal \\n in input. I'm a little fuzzy on how you can ever get a \n out of your history commands, though.

I tried w/ zsh to find out how a \newline might be read out from the history file without being escaped like that:

echo '
\n'
history -1

---

11355 echo '\n\n'

That won't do. So, I tried this instead:

hist(){ ! cat "$1"; }
echo '
\n'
FCEDIT=hist fc -1

---

echo '
\n'

...so that's probably what you should be doing instead

For some reason zsh escapes \newlines in history or fc -l output in an ambiguous way, but when it hands a history command over to some editor, it gives it the real thing.

zsh's primary means of history manipulation is fc, and the history command is not much more than an alias for fc -l. When fc is called with the -l option it will list the matching history lines to stdout (after ambiguously escaping any non-printable characters), but fc's default behavior is to invoke another utility with one or more temp file arguments into which it has written all history matches for its args.

fc derives the name of the utility it invokes from the $FCEDIT environment variable, or, if it is unset or null, from $EDITOR, or, if likewise unsuccessful, defaults to vi. If the invoked utility returns true, fc will afterward attempt to run any commands it finds in the (presumably) edited temp files before removing them.

And so the above command sequence substitutes cat for any editor command, and inverts its return so that a successful readout of fc's tempfiles will return false - to keep fc from trying to run the commands again.

A more complete, sort-of drop-in solution which doesn't need to call out to an external utility could look like:

history()
        case    $1      in
        (*/*) ! while   [ -r "$1" ]
                do      while   read -rE RTN
                        do :;   done    <$1
                shift;  done;;
        (*)     local   RTN=1  IFS=
                FCEDIT=history fc "$@"
                return  $((RTN*$?))
        esac

...which ought to handle almost transparently any argument list you might expect zsh's history command to do; except that its output is always literal and it never includes history timestamps or event numbers in the output (unless you call it like history -l [args], in which case it will behave as the builtin history).