Feature idea: Expanding on completion functions

Feature idea: Expanding on completion functions

[Richard Hartmann]

Hi all,

quite often, I wish I could use the output of a completion function
on the comand line. Think

  my_list=$(magic_function(apt-get install openoffice-<tab>))

The completion functions are so powerful it's a shame not to use
their output for more purposes.

Bart(?) told me this would not be an easy thing to do so I think
it makes sense to implement this through a shell feature or not
at all.


Thoughts, feedback, suggestions?
Richard

Re: Feature idea: Expanding on completion functions

[Bart Schaefer]

On Nov 30, 10:35pm, Richard Hartmann wrote:
}
} The completion functions are so powerful it's a shame not to use
} their output for more purposes.

Without necessarily shooting down this idea, let me point out that
completion functions don't have output.  I think the following is
mostly similar to what I said before:

What completion functions do is shove a large number of *possible*
completions into the "compadd" builtin.  Typically they generate WAY
more possibilities than you ever see as "output" ... a lot of the
magic hides in the way completion uses the contents of the command
line at or around the cursor position to throw away most of the data
that was fed to it, so you see the relevant remainder.

The rest of the magic comes from the constellation of variables that
describe the command line context so that the completion functions
have some idea where to look for possible completions.

To use the completion functions for more purposes, you have to both
populate all those context variables appropriately to each of those
other purposes, and also create the appropriate filters to remove
whatever is irrelevant to each purpose.

And then you have to decouple the stuff that performs the filtering
from the stuff that displays the results in ZLE, which is the only
place that any "output" goes right now.

-- 

Re: Feature idea: Expanding on completion functions

[Richard Hartmann]

On Sun, Nov 30, 2008 at 23:23, Bart Schaefer <schaefer@brasslantern.com> wrote:

> Without necessarily shooting down this idea, let me point out that
> completion functions don't have output.

No, no, feedback is good. Thanks for taking the time to write it out
in this level of detail.


> And then you have to decouple the stuff that performs the filtering
> from the stuff that displays the results in ZLE, which is the only
> place that any "output" goes right now.

Wouldn't a relatively easy way to do this be to copy all state and
execute the completion in this context? All the filtering etc could
happen in the normal way as all potentional output which is filtered
away is not interesting (to me) at that point, anyway.


Richard