Re: inputting characters by hexadigit

[Ted Zlatanov]

On Wed, 23 Jul 2008 15:31:36 -0400 Stefan Monnier <address@hidden> wrote: 

>> Can completions be cascaded somehow?  The first tier would show all the
>> common first words, e.g. ... AEGEAN APL GREEK ... and then selecting
>> something from the first tier would cascade down to the second tier.

SM> The slow display should only happen when the list is really long,
SM> i.e. basically if you hit TAB with an empty minibuffer.  So we could
SM> indeed easily use a different completion behavior in this case.

>> This might actually be a sensible suggestion for any completion list
>> with more than 100 or so members, and the tier separation can be done
>> with the word syntax of the current mode.

SM> You mean show the list as a tree?  That cannot be done with the current
SM> default completio code.  But some variant of it can be done.  E.g.  we
SM> can use a completion similar to file names, so additionally to
SM> <charname> we could accept input of the form <category>/<charname> and
SM> provide completion on this (and if the completion list on <charname> is too
SM> long, don't display it: only display the completion on
SM> <category>/<charname>).

To clarify my proposal:

Split on the word syntax, so (taking the Unicode example) hitting TAB
initially gives

APL (200 candidates)
AEGEAN (20 candidates)
GREEK (50 candidtates)

then as the user types A, the list becomes

APL (200 candidates)
AEGEAN (20 candidates)

with the A highlighted, and so on.  As soon as less than X candidates
are available the usual list is displayed; X is customizable but 40 by
default.

The word syntax can depend on the current mode, or be overridden by the
function being called.

The advantage over categories, which you suggested, is that the
completion code won't need categories supplied--it automatically gets
them based on the word syntax, so it's less work for the programmer and
thus more likely to get used everywhere and be suitable as a drop-in
replacement for the current completion code.  The current behavior
(e.g. M-x TAB) is nearly unusable without either apropos or icicles
(that I know of) to search through the list.

Ted