Ruby readline - completion based on already-entered commands

How can I use the builtin readline library to complete arguments to
commands? I have basic completion working now, so that

%wmi > st [TAB]

will show valid commands starting with "st" (in this case "start" and
"stop"), but how do I complete arguments specific to that command?

For instance, if the user has already typed "start Te" and then hits
TAB, it should show completions specific to the start command (in this
case "TestDevice" and "TerminalServer"), but I'm not clear on how to
do this.

···

--
Bill Atkins

Do you have a Pickaxe (second edition)? It's demoed on the page for the Readline library.

Let me know if you don't and I'll post an example...

James Edward Gray II

···

On Aug 11, 2005, at 9:44 AM, Bill Atkins wrote:

How can I use the builtin readline library to complete arguments to
commands? I have basic completion working now, so that

%wmi > st [TAB]

will show valid commands starting with "st" (in this case "start" and
"stop"), but how do I complete arguments specific to that command?

Unfortunately the ruby readline bindings are missing some key functions to make this easy. Even raw access to the line buffer would make this trivial, but this is not available.

I used the following code to augment the ruby library 'Cmd' (http://code.vernix.org/cmd\) to implement what you are asking for (the missing functions in the code below will be found in this library, but they just do what their name implies). The main problem to solve is how to gain access to the whole command line and not just the last token being completed (this is all readline give you with the Ruby API). The Ruby API doesn't provide this directly but you can change the word split character to be an unused character (I use bell character). This gets you the complete line to make your completion decision on, but you need to also include the earlier characters on the line in the completion list you pass back to readline. This unfortunately causes <tab><tab> to prepend the whole line on every possible completion in the presented completion list.

You also have to worry about completing on an empty set inserting spaces into the line buffer (more cosmetic than anything else if I recall). The space is fixed by turning off the auto space on complete in readline and adding this yourself. I also output the bell when necessary. Still not perfect behaviour as <tab> <tab> on an empty set should just beep but not display anything while in this case it will display the line so far.

       # Configure readline
       set_completion_proc(:readline_completion_handler)
       Readline.completer_word_break_characters = 7.chr # bell character
       Readline.completion_case_fold = true
       # Our empty completion sets return the original line so readline thinks we have sucessfully completed
       # and will append the following character. We need to stop it from doing this.
       Readline.completion_append_character = ''

     def readline_completion_handler(line)
         tokens = line.split
         tokens << "" if line =~ /^\w.* $/ # starting new completion set
         if tokens.length == 0
             command_list
         else
             # Get completion set for token.last. See if we have a valid single command with a completion proc defined.
             completion_proc = nil
             commands = completion_grep(command_list, tokens.first)
             if commands.size == 1
                 cmd = commands.first
                 completion_proc = complete_method(cmd) if collect_complete.include?(cmd)
             end
             if tokens.length == 1
                 matches = completion_grep(command_list, tokens.first)
                 matches = matches.map {|m| m + ' '} if matches.length == 1
                 matches
             else
                 if completion_proc
                     tokens.shift # don't need to send the command
                     matches = self.send(completion_proc, tokens)
                     if matches.empty?
                         print "\a" # beep
                         [line] # empty completion set so pass back original line
                     else
                         # Prepend the command line up to the last token we have been completing on to all the match values.
                         line.sub!(Regexp.new("#{tokens.last}$"), "")
                         matches = matches.map {|m| line + m}

                         # If match is complete add space to move onto next token
                         matches = matches.map {|m| m + ' '} if matches.length == 1
                         matches
                     end
                 else
                     print "\a" # beep
                     [line] # empty completion set so pass back original line
                 end
             end
         end
     end

Hope this helps,

Dave.

···

On 11 Aug 2005, at 15:44, Bill Atkins wrote:

How can I use the builtin readline library to complete arguments to
commands? I have basic completion working now, so that

%wmi > st [TAB]

will show valid commands starting with "st" (in this case "start" and
"stop"), but how do I complete arguments specific to that command?

For instance, if the user has already typed "start Te" and then hits
TAB, it should show completions specific to the start command (in this
case "TestDevice" and "TerminalServer"), but I'm not clear on how to
do this.

-- Bill Atkins

I do. I have the basics of readline working; I'm just trying to
figure out how to complete input differently based on already entered
commands.

An analogy would be shell completion. You type "mpla [TAB]" and
"mplayer" gets completed, but any subsequent completion is for paths
only.

···

On 8/11/05, James Edward Gray II <james@grayproductions.net> wrote:

On Aug 11, 2005, at 9:44 AM, Bill Atkins wrote:

> How can I use the builtin readline library to complete arguments to
> commands? I have basic completion working now, so that
>
> %wmi > st [TAB]
>
> will show valid commands starting with "st" (in this case "start" and
> "stop"), but how do I complete arguments specific to that command?

Do you have a Pickaxe (second edition)? It's demoed on the page for
the Readline library.

Let me know if you don't and I'll post an example...

James Edward Gray II

--
Bill Atkins

Oops, sorry. I misunderstood the question... and it's a great one! I too await the answer...

James Edward Gray II

···

On Aug 11, 2005, at 10:00 AM, Bill Atkins wrote:

An analogy would be shell completion. You type "mpla [TAB]" and
"mplayer" gets completed, but any subsequent completion is for paths
only.

You'd have to have the completion function parse the line "so far" so you'd have it match against commands until there was only one command to match ("mplayer") then match completions against file names, returning options concatenating the command and file name options.

-Brian

···

On Aug 11, 2005, at 11:00 AM, Bill Atkins wrote:

I do. I have the basics of readline working; I'm just trying to
figure out how to complete input differently based on already entered
commands.

An analogy would be shell completion. You type "mpla [TAB]" and
"mplayer" gets completed, but any subsequent completion is for paths
only.

On 8/11/05, James Edward Gray II <james@grayproductions.net> wrote:

On Aug 11, 2005, at 9:44 AM, Bill Atkins wrote:

How can I use the builtin readline library to complete arguments to
commands? I have basic completion working now, so that

%wmi > st [TAB]

will show valid commands starting with "st" (in this case "start" and
"stop"), but how do I complete arguments specific to that command?

Do you have a Pickaxe (second edition)? It's demoed on the page for
the Readline library.

Let me know if you don't and I'll post an example...

James Edward Gray II

--
Bill Atkins

No problem.

···

On 8/11/05, James Edward Gray II <james@grayproductions.net> wrote:

On Aug 11, 2005, at 10:00 AM, Bill Atkins wrote:

> An analogy would be shell completion. You type "mpla [TAB]" and
> "mplayer" gets completed, but any subsequent completion is for paths
> only.

Oops, sorry. I misunderstood the question... and it's a great one!
I too await the answer...

James Edward Gray II

--
Bill Atkins