Strip Filenames and Rename

I believe that this program will do what you want. I have tested it and it works on several interesting cases (such as no extension at all):

#!/bin/bash

for fname in *; do
  name="${fname%\.*}"
  extension="${fname#$name}"
  newname="${name//./_}"
  newfname="$newname""$extension"
  if [ "$fname" != "$newfname" ]; then
    echo mv "$fname" "$newfname"
    #mv "$fname" "$newfname"
  fi
done

The main issue you had was that the ## expansion wasn't doing what you wanted. I've always considered shell parameter expansion in bash to be something of a black art. The explanations in the manual are not completely clear, and they lack any supporting examples of how the expansion is supposed to work. They're also rather cryptic.

Personally, I would've written a small script in sed that fiddled the name the way I wanted, or written a small script in perl that just did the whole thing. One of the other people who answered took that approach.

One other thing I would like to point out is my use of quoting. Every time I do anything with shell scripts I remind people to be very careful with their quoting. A huge source of problems in shell scripts is the shell interpreting things it's not supposed to. And the quoting rules are far from obvious. I believe this shell script is free of quoting issues.

With zsh:

autoload zmv
zmv '(**/)(*.*)' '$1${2//./_}'

Otherwise, if you have access to bash (though ksh93 or zsh will do as well), you could always do:

find . -depth ! -name '.*' -name '*.*' -exec bash -c '
  for file do
    dir=${file%/*}
    base=${file##*/}
    mv -i -- "$file" "$dir/${base//./_}"
  done' sh {} +

(though you'll miss the sanity checks done by zmv).

Those (intentionally) don't rename hidden files.

The point is to process the list depth-first (which zmv does by default, and using -depth in find) and only rename the basename of the file.

So that you do:

mv -- a.b/c.d a.b/c_d

and then:

mv -- a.b a_b

Also note that for the non-zmv solution, if you have a a.b file and a a_b directory in the same directory, then mv -- a.b a_b will happily move a.b into a_b/. To avoid that, if on a GNU system, you can add the -T option to mv.

As I see you tried to edit the answer to use a while read loop.

Note that you should generally avoid while read loops, even in this case where one command per line has to be run anyway. If you really want to use one, then doing it right is tricky and involves ksh/bash/zshisms:

{
  find . -depth ! -name '.*' -name '*.*' -exec printf '%s\0' {} + |
    while IFS= read -rd '' file; do
      dir=${file%/*}
      base=${file##*/}
      mv -i -- "$file" "$dir/${base//./_}"
    done <&3 3<&-
} 3<&0

(you can replace -exec printf '%s\0' {} + with -print0 if your find supports it).

You need:

• -print0/-exec printf '%s\0' {} + as you can't rely on newline as the separator since newline is a valid character in a filename
• as a result, you need -d '' (read reads until NUL instead of NL)
• IFS= (that is remove all the whitespace characters from IFS for read as otherwise strips them from the end of the input record).
• -r as otherwise read treats backslash as an escape character (for the field and record separator).
• that litte dance with file descriptors because you want the standard input of mv to be the terminal (for the answers to the -i prompts), not the pipe from find.