If you want the equivalent of
sed -i.bak, it's pretty simple.
Consider this script for GNU sed:
# Create an input file to demonstrate
trap 'rm -r "$dir"' EXIT
grep -v '[[:upper:][:punct:]]' /usr/share/dict/words | head >"$dir/foo"
# sed program - removes 'aardvark' and 'aardvarks'
# What we want to do
sed -i.bak -e "$script" "$dir"
# Prove that it worked
We can simply replace the marked line with
cp "$dir/foo" "$dir/foo.bak" && sed -e "$script" "$dir/foo.bak" >"$dir/foo"
This moves the existing file to be a backup, and writes a new file.
If we want the equivalent of
sed -i -e "$script" "$dir" # no backup
then it's slightly more complex. We can open the file for reading as standard input, then unlink it, before directing sed's output to replace it:
( cp "$dir/foo" "$dir/foo.bak"; exec <"$dir/foo.bak"; rm "$dir/foo.bak"; exec sed -e "$script" >"$dir/foo" )
We do this in a sub-shell, so that our original stdin is still available after this. It's possible to switch inputs and switch back without a subshell, but this way seems clearer to me.
Note that we're careful to copy first, rather than creating a new
foo file - this is important if the file is known by more than one name (i.e. has hard links) and you want to be sure that you don't break the links.