BUG: Extended completion with alternative completion

["Bart Schaefer" <schaefer@xxxxxxxxxxxxxxxx>]

I'm fooling with the compctl for cdmatch from Misc/compctl-examples:

compctl -K cdmatch -S '/' -q -x 'p[2]' -Q -K cdmatch2 - \
	'S[/][~][./][../]' -g '*(-/)' + -g '*(-/D)' - \
	'n[-1,/]' -K cdmatch -S '/' -q -- cd chdir pushd

With this compctl installed (and no others except the 3.1.5 defaults, which
are essentially none), from the zsh-3.1.5 directory, the following setopts:

noautomenu            off
completealiases       off
completeinword        on		<-- This is important
globcomplete          off
menucomplete          off

I type:

zagzig<7> cd S/M

Now back up so the cursor is on the `S' and press TAB.  The n[-1,/] case is
chosen correctly by get_ccompctl() and returned to makecomplist().  At this
point we hit the following code:

    /* *compadd is the number of characters we have to ignore at the *
     * beginning of the word.                                        */
    wb += *compadd;
    s += *compadd;
    if ((offs -= *compadd) < 0)
	/* It's bigger than our word prefix, so we can't help here... */
	return 1;

offs <= *compadd, so 1 is returned to docompletion() at this code:

            /* Make sure we have the completion list and compctl. */
            if(makecomplist(s, incmd, &delit, &compadd, untokenized)) {
                /* Error condition: feeeeeeeeeeeeep(). */
                feep();
                goto compend;
            }

At this point we feep() and fail, mistakenly believing that an error has
occurred.  This appears harmless until the compctl is modified like so:

compctl -K cdmatch -S '/' -q -x 'p[2]' -Q -K cdmatch2 - \
	'S[/][~][./][../]' -g '*(-/)' + -g '*(-/D)' - \
	'n[-1,/]' -K cdmatch -S '/' -q -- + -K cdmatch -S '/' -q \
	cd chdir pushd                   ^^^^^^^^^^^^^^^^^^^^^^^

There are two problems: (1) in get_ccompctl(), when we chose to return
the n[] completion, we lose track of the alternative (which hangs off
the xor pointer in the head of the linked list of -x patterns); (2) we're
prematurely returning an error in makecomplist(), so even if we still had
the handle to the xor, we'd never follow it.

In short, alternative completion only works with extended completion when
none of the extended patterns match, which doesn't seem right to me.

It's not hard to fix (2), but I'm leery of messing with (1) when there are
a lot of other pending completion patches to be folded in to 3.1.6.  Sven,
are you out there?

-- 
Bart Schaefer                                 Brass Lantern Enterprises
http://www.well.com/user/barts              http://www.brasslantern.com