Re: How to avoid expansion when completing?

[Sven Wischnowsky <wischnow@informatik.hu-berlin.de>]

Peter Stephenson wrote:

> > Suppose I have a load of subdirectories, each of which contains a file
> > called foobar.dat.  I enter 
> > 
> > % wc **/foob<Tab>
> > 
> > hoping to expand this to **/foobar.dat (assuming there's nothing else
> > starting with 'foob' in any of the subdirectories).  However, I get the
> > '**' expanded to the first subdirectory name which has a file starting
> > with 'foob' in it.  (If I keep pressing <Tab> then I cycle round).
> > 
> > I've got the following options set, and I'm running version 3.1.5.
> 
> If you don't mind upgrading, you can do such things with the new bindable
> completion widgets in 3.1.9 (maybe 3.1.6, but that's irrelevant if you're
> still on 3.1.5).  Assuming the file is unique, you can use
> e.g. _most_recent_file (usually bound to ^xm), which would normally put a
> `*' on the end and find the most recently modified matching file --- that
> ought to be enough for this purpose.
> 
> The following function is a trivial modification of that to add all
> matching files as possible completions (_glob_expand_and_complete).
> Sven probably has a better way of doing it.

Actually, no. I seem to remember Andrej the same question. And I seem
to remember me replying that this could get pretty expensive...

One could write a function that globs and lets one complete the last
pathname component. But it would then probably be easy to find cases
where it should complete the `intersting' part instead, which can be
hard (or even impossible) to identify with complex patterns. Especially
if the possible completions can have different numbers of components.

After thinking about this I gave up thinking about it at all.

It might be possible to add this to _match (together with _path_files;
for those not familiar with the new completion system: these are two
of its functions, _path_files completes filenames and _match offers
something like globcomplete, opnly more powerful). In fact, they
already handle completion of patterns a bit, but they don't try to
restore parts of the completion back to the original pattern, which
would be needed here.

Bye
 Sven


--
Sven Wischnowsky                         wischnow@informatik.hu-berlin.de