RE: [External] : Re: [PATCH v3] Allow applying filters to summary consecutively

[Drew Adams]

> Here's an idea.  Type & to do intersection with the following filter,
> and type | to do union with the following filter.  Both & and | are
> undefined in Rmail mode, so they are available for this.  They
> will be easy to remember, too.

FWIW, this is what you can do using Icicles - HTH.

The idea is that (1) you can narrow to a subset of
the current set of candidates, or (2) if you've
narrowed, you can widen to the current set plus
new matches against the last set from which you
narrowed.

IOW, you can widen if you've previously narrowed
- otherwise, widening is a no-op, as there's no
superset to match against.

It works like this:

1. When you complete, the original domain of
completion candidates is that used by (passed to)
`completing-read' (or similar) - an alist or whatever.

2. During completion you can use a key to narrow or
widen the set of candidates.

3. Each time you narrow, a recursive minibuffer is
used, completing with the result of narrowing, i.e.,
completing against the set of matches in the parent
minibuffer level.

4. `C-g' pops up one level of minibuffer, i.e., one
level of recursion (out of the minibuffer altogether
if already at the top-level minibuffer).

5. #4 means you can narrow, do something with some
candidates, then `C-g' to remove the last narrowing
filter, narrow again with a different filter, etc.

6. When you widen, the current set of candidates is
matched against a regexp that's an alternation (\|)
of the current match pattern and whatever widening
pattern you provide.  Instead of entering a recursive
minibuffer, you stay at the same level, but with the
match-pattern changed to the alternation.

Narrowing matches patterns independently - IOW it's
orderless.  This is possible because it uses a fresh
minibuffer with a different set of candidates (those
that matched the previous pattern of the new AND).

Widening just changes the current match-pattern to
an alternation regexp (OR).  It doesn't invoke a
recursive minibuffer.  It's not orderless.
____

`M-*' narrows without completing
`S-SPC' narrows and completes
`M-+' widens without completing
`S-<backspace>' widens and completes
`S-TAB' completes (`TAB' does also, but differently)