How do I pipe `dos2unix` into `while read` in bash?

Question

Right now, I'm trying to fix an issue in my PrintBans.sh script.

The problem is, the program that generates this file saves it with \r\n line endings, so I need the while loop to be able to read \r\n lines, otherwise there's an extra \r at the end of the last line which results in the arithmetic failing:

 - 621355968000000000")syntax error: invalid arithmetic operator (error token is "

I've tried these.

while read ban
do
    ...
done < dos2unix $file

while read ban
do
    ...
done < `dos2unix $file`

cat $file > dos2unix > while read ban
do
    ...
done

while read ban
do
    ...
done < dos2unix < $file

I also see that some people set IFS='\r\n', but this did not work for me.

Is it impossible to pipe files through dos2unix without overwriting the original file?

Answer 1

Literal Answer: Pipe Through!

If you don't tell dos2unix the name of the file it's working with, it can't modify that file in-place.

while IFS= read -r line; do
  echo "No carriage returns here: <$line>"
done < <(dos2unix <"$file")

Redirections are performed by the shell before a program is started, when you invoke dos2unix <input.txt, the shell replaces file descriptor 0 with a read handle on input.txt before invoking dos2unix with no arguments.

_{If you wanted to be really paranoid (and pay a performance cost for that paranoia), you could prevent a hypothetical nonexistent dos2unix that modified a file descriptor received on stdin in-place from doing so by making it <(cat <"$file" | dos2unix), such that dos2unix is reading from a FIFO connected to the separate executable cat, rather than straight from the input file. Needless to say, I don't ever advise this in practice.}

---

Better Answer: Don't

You don't need dos2unix (which -- with its default in-place modification behavior -- is meant for human interactive users, not scripts); the shell itself can strip carriage returns for you:

#!/usr/bin/env bash
#              ^^^^- not /bin/sh; needed for $'' syntax and [[ ]]

while IFS= read -r line || [[ $line ]]; do
  line=${line%$'\r'}
  echo "No carriage returns here: <$line>"
done <"$file"

• ${var%expr} is a parameter expansion which strips any trailing instance of the glob expression expr from the contents of the variable var.

• $'\r' is ANSI C-like string syntax for a carriage return. Using that syntax is important, because other things that look like they might refer to a carriage return don't.

  • \r outside any kind of quoting context is just the letter r.
  • "\r" or '\r' are two characters (a backslash and then the letter r), not a single carriage return.
• [[ $line ]] is a ksh extension adopted by bash equivalent to [ -n "$line" ]; it checks whether the variable line is non-empty. It's possible for read to return false while still populating a line if you have a partial line without any terminator; as l0b0 points out, line separators rather than terminators are common on Windows. This ensures the last line of a file is processed even if it doesn't end in a newline.