Is it possible to configure the way bash completes directory names?

I'd like to instruct bash to use a special method to perform completion on certain directory names. For example, bash would call a program of mine to perform completion if a path starts with "$$", and perform completion normally otherwise.

Is this at all possible? How would you implement it?

Bounty: I'd really appreciate an answer to that question. The goal is to allow autojump to complete paths for all commands when the user starts them with a certain prefix. So for example when copying a file from a far directory, you could type:

cp $$patern + <Tab>

and autojump would complete

cp /home/user/CompliCatedDireCTOry/long/path/bla/bla

and you would just have to add where you want to put the file. Of course I can use ott's comment to add it to a few specific commands, but if anyone has a better idea, I'd be very grateful.

You can do this by overriding the default binding for TAB(^i). First you need to override the TAB binding, then you need to build a function that calls your command, lastly you need to take the output from that command and update the variable that contains the current command line.

This function takes the current command line and changes the last two characters to 'huugs'.

function my_awesome_tab_completion_function () {
  set -- $READLINE_LINE
  command="$1"
  shift
  argument="$*"
  argument_length=$(echo -n $argument | wc -c)
  if echo $argument | grep '^$$' >/dev/null 2>&1; then
    new_argument=$(echo $argument | sed 's/..$/huugs/') # put your autojump here
  else
    new_argument=$(compgen -d $argument)
  fi
  new_argument_length=$(echo -n $new_argument | wc -c)
  READLINE_POINT=$(( $new_argument_length - $argument_length + $READLINE_POINT ))
  READLINE_LINE="$command $new_argument"
}

For your example you'd probably want to change the new_argument line to look like this:

  new_argument=$(autojump $argument)

Now override the ^i binding:

$ bind -x '"\C-i"':'my_awesome_tab_completion_function'

Now test that it works:

$ cd /ro<TAB>
changes my command to:
$ cd /root

so normal completion still works, you can test the $$ part by doing cd $$... etc

If you run into issues turn on verbose mode:

$ set -x

It will print out everything the function is doing.

I tested this on Ubuntu 11 using bash 4.2.8(1)-release (the default).

The bash completion routine can be programmed as a shell script.

Here is an example of a shell-script that will replace in any parameter $$[Tab] by my replacement string, but only for the specific command mycommand and only if the parameter is exactly "$$" :

_mycomplete()
{
        if [ ${COMP_WORDS[COMP_CWORD]} == \$\$ ]
        then
                COMPREPLY='my replacement string'
        fi
}
complete -o default -o bashdefault -F _mycomplete mycommand

You must source the script to bash via source <file-name> (or the dot command) to start it working, and then :

mycommand $$[Tab] -> mycommand my replacement string
mycommand $$$[Tab] -> mycommand $$$ (beep)
mycommand whatever[Tab] -> (will complete "whatever" in the normal bash manner)

To always have it working for some or all users, include this in one of the bash profile routines.

The problem with the complete command, is that it will work only for one or more names of commands, that are specified as parameters. One could simply give it the list of all the commands that could possibly be used by the users, or in desperate cases expand /bin/* /usr/bin/* ~/bin/*.

Tested on CentOS 5.5.

_{This simple script is based on the sources I have listed in my other answer - the one which was deleted by the moderator studiohack. If interested, just ask him to undelete it.}