Can Bash achieve this kind of completion ?

Question

I'm a bash user, who's been checking out a few zsh configs lately, to see what it's made of.
And for the most part, I haven't seen many things that I couldn't live without, except for this nifty completion thingy :
This config provides a completion mechanism, that displays little docstrings explaining a bin's options, which is so damn useful.
When you type something like :
➜ ~ grep -[TAB]
You get :
--after-context -A -- specify lines of trailing context
--basic-regexp -G -- use basic regular expression
--before-context -B -- specify lines of leading context
--binary-files -- specify type to assume for binary files
--byte-offset -b -- print the byte offset with output lines
--colour --color -- distinguish matching string
--context -C -- specify lines of context
--count -c -- only print a count of matching lines
--devices -D -- specify handling of devices, FIFOs and sockets
--directories -d -- specify handling of directories
--exclude -- skip files matching specified pattern
--exclude-from -- skip files matching pattern in specified file
--extended-regexp -E -- use extended regular expression
--file -f -- specify pattern file
--files-with-matches -l -- output matching files' names only
--files-without-match -L -- output non-matching files' names only
--fixed-strings -F -- use literal strings
--help -- display help
-I -- process binary files as if non-matching
--ignore-case -y -i -- case-insensitive
--include -- examine files matching specified pattern
--invert-match -v -- select non-matching lines
--label -- provide filename to print for stdin
--line-buffered -- flush output on every line
--line-number -n -- prefix output with line numbers
--line-regexp -x -- force pattern to match only whole lines
--max-count -m -- stop after specified no of matches
--mmap -- memory map input
--no-filename -h -- suppress printing of filenames
--no-messages -s -- suppress messages about unreadable
--null -Z -- print 0 byte after FILE name
--only-matching -o -- show only matching part of line
--perl-regexp -P -- use perl regular expression
--recursive -R -r -- recurse subdirectories
--regexp -e -- specify pattern
--silent --quiet -q -- suppress normal output
--text -a -- process binary file as if it were text
--version -V -- display version info
--with-filename -H -- print filename with each match
--word-regexp -w -- force pattern to match only whole words
That's just awesome, and I really lust for this feature now...
I haven't made the jump to zsh, as I'd rather stick with stuff that's available on virtually every machine I use (zsh users don't flame, I know it's basically better than bash, I'd just rather have portability over functionality most of the time).
So any bash gurus know if this can be done ?

Answer 1

I've been looking for that also... I think there is no "good" way to do it, but I use an ugly trick to get something similar.
For example, for the kill command I use the following "complete" rule
_kill ()
local cur aux
COMPREPLY=()
cur=${COMP_WORDS[COMP_CWORD]}
COMPREPLY=( $( compgen -W '$( ps -U$USER -o pid )' -- "$cur" ))
if [ $COMP_TYPE -eq 63 ]; then
echo ""
ps -lU$USER |\
awk -v cur=$cur '{if (index($4,cur)==1 || NR==1)\
print "\033[1m" $4 "\033[0m" "\t" $5 "\t" $14}'
fi
return 0
complete -F _kill kill
which gives something like this
olvar:~$ kill 10[TAB][TAB]
PID PPID CMD
10397 1921 urxvt
10405 10397 bash
10397 10405
olvar:~$ kill 10
Of course this is ugly, it uses ps twice and sorts the results twice, not to mention that you have the matches repeated in the screen,
but I haven't found anything better nor how to hide the $COMPREPLY array.
For grep i think I'd do something like
_grep ()
local cur
cur=${COMP_WORDS[COMP_CWORD]}
if [[ "$cur" = -* && $COMP_TYPE -eq 63 ]]; then
echo ""
$1 --help 2>&1
echo ""
fi
_longopt $@
complete -o filenames -F _grep grep
Probably you would like to parse the output better than just the "$1 --help 2>&1" I wrote.
Anyone else with a better suggestion?
Last edited by olvar (2010-03-11 20:32:52)