Magic-Backspace

Magic-Backspace

[Pascal Wittmann]

[-- Attachment #1: Type: text/plain, Size: 478 bytes --]

Hi,

I thought of implementing some magic-backspace that removes characters
until the 'current' completion is ambiguous (i.e. some extended version
of undo).

For example, if I type

	autoconf

and press 'magic-backspace' it would result in

	auto

The problem I'm facing is that I can not find out if the current
completion is ambiguous. Is this possible at all?

(I've tried to abuse complete-word and friends, but without success).


Regards
Pascal



[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 198 bytes --]

Re: Magic-Backspace

[Bart Schaefer]

On Sep 11,  9:51am, Pascal Wittmann wrote:
}
} I thought of implementing some magic-backspace that removes characters
} until the 'current' completion is ambiguous (i.e. some extended version
} of undo).
} 
} The problem I'm facing is that I can not find out if the current
} completion is ambiguous. Is this possible at all?

This is a rather tricky thing to attempt because the completion system
is a bit more than just a ZLE widget.  Completion wants to have control
of the editing process, i.e., you get *either* direct manipulation of
the editor buffer with BUFFER, LBUFFER, RBUFFER, etc., *or* you get the
completion system mucking about in the buffer as it deems necessary,
but not both at once.

That's one reason why completion widgets are created with "zle -C ..."
rather than with "zle -N ...".

So the best approach I can think of is to do this in two parts:  an
ordinary zle widget, that is a wrapper around a completion widget.  The
wrapper widget declares some variables as locals which the completion
widget then uses but does not declare.

This new completion widget does nothing but call _main_complete, then
copy some values from the $compstate hash into the shared variables,
and finally assign compstate[insert]='' so that the command line will
not be changed.  You need $compstate[unambiguous], and possibly also
context, unambiguous_cursor, unambiguous_positions, insert_positions.

Your wrapper will call the completion widget with "zle widget-name"
and then examine the shared variables to deterimine whether to delete
a character.  As you can probably tell, the completion system is very
strongly tied to detecting when the line is NOT ambiguous, so you're
going to have to figure out how to extrapolate back from that.

Obviously there's a whole lot of stuff I've glossed over here.  What
should your widget do if invoked in the middle of a word instead of
at the end?  What state do you need to clear before, or set after,
calling the completion widget to avoid entering menu completion or
displaying a listing?  And so on.

Have fun!

Re: Magic-Backspace

[Pascal Wittmann]

[-- Attachment #1: Type: text/plain, Size: 1219 bytes --]

On 09/11/2012 05:09 PM, Bart Schaefer wrote:
> On Sep 11,  9:51am, Pascal Wittmann wrote:
> }
> } I thought of implementing some magic-backspace that removes characters
> } until the 'current' completion is ambiguous (i.e. some extended version
> } of undo).
> } 
> } The problem I'm facing is that I can not find out if the current
> } completion is ambiguous. Is this possible at all?
> 
> This is a rather tricky thing to attempt because the completion system
> is a bit more than just a ZLE widget.  Completion wants to have control
> of the editing process, i.e., you get *either* direct manipulation of
> the editor buffer with BUFFER, LBUFFER, RBUFFER, etc., *or* you get the
> completion system mucking about in the buffer as it deems necessary,
> but not both at once.
> 

 [snip]

> 
> Obviously there's a whole lot of stuff I've glossed over here.  What
> should your widget do if invoked in the middle of a word instead of
> at the end?  What state do you need to clear before, or set after,
> calling the completion widget to avoid entering menu completion or
> displaying a listing?  And so on.
> 
> Have fun!
> 

Thanks for all these pointers! Now I've a starting point.


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 198 bytes --]

Re: Magic-Backspace

[Oliver Kiddle]

On 11 Sep, Pascal Wittmann wrote:
> 
> I thought of implementing some magic-backspace that removes characters
> until the 'current' completion is ambiguous (i.e. some extended version
> of undo).

I've got a function which does roughly this. It has a number of issues
and, as I wrote it over eight years ago, I can't remember what those
issues are. I've attached it below. It works by removing characters
from the PREFIX and trying completion. It is intended to be used as a
completion widget via _generic:
  zle -C retract complete-word _generic
  zstyle ':completion:retract::::' completer _retract
  bindkey '^[[24~' retract  # F12

I'm not sure what the commented out line was for.

Oliver

#autoload

local _comp_backup=1 bumax=30

if (( ! $+functions[compadd] )); then
  compadd() {
    local save oldp="$PREFIX"
    SUFFIX=
    # removing the -O option is a nasty way to make this work from _multi_parts
    zparseopts -D -E O:=save
    local i
    for ((i=_comp_backup;i>0;i--)); do
      PREFIX="${PREFIX%?}"
    done

    builtin compadd -S '' -F "( $oldp* )" "$@"
  }
  trap 'unfunction compadd' EXIT INT
fi

while [[ _comp_backup -le bumax ]]; do
  _complete && return
  (( _comp_backup++ ))
done

#builtin compadd - ''

return 1