Re: A dynamic pattern-matching

[Francois Fleuret]

Hi Stefan,

 > - it'd be easy to write a variant of icomplete-mode that addresses
 >   point 1 by using a separate buffer (which we'd call
 >   "*Completions*").

Well, that's not clear to me. IIUC completing-read and other functions
of that kind blocks the calling program until the user makes up his
mind, while my selector/select uses a callback.

In fact, is it really desirable to use that selector in place of the
standard completion functions ? My idea was more to provide an
alternative to recentf-open-files and switch-buffer.

 > - I guess point 2 is also somewhat secondary, tho it's important to
 >   have a notion of "the currently selected entry" (this is also
 >   used in ido/ibuffer and is somewhat missing in the default
 >   completion code)

In fact, I always considered using a "real buffer" to be a major
(good) thing, but now that you talk about it I wonder why, since what
one usually does in the selector buffer is not interrupted by
something else. Now I presume it's a question of habit, but I never
felt really at home in the minibuffer ...

 > - I don't think point 3 is important, right?

Nope, it's not

 > - point 4 can be simulated by adding a "*" at the beginning of
 >   the pattern.  Or we could add a `substring' completion-style.

Yes. But I would vote for having a variable that specifies how the
typed pattern has to be interpreted { substrings, regexp,
leadingstring } or something.

 > - point 5 hopefully can be implemented as a completion-style.

Not sure to understand what you do mean here.

 > - point 6... not sure: it seems like the difference might be important,
 >   along with the fact that the user may like to use this style of
 >   selection for some commands but not all.

See my comment about the 1/

Regards,

-- 
Francois Fleuret                            http://www.idiap.ch/~fleuret/