\## -> x when completing

\## -> x when completing

[Mikael Magnusson]

zsh -f
touch xenon
autoload compinit; compinit
setopt globcomplete
cat \##<tab> -> cat xenon

-- 
Mikael Magnusson

Re: \## -> x when completing

[Bart Schaefer]

This gets a bit stream-of-consciousness but maybe that's helpful.

On Feb 8, 11:37pm, Mikael Magnusson wrote:
} Subject: \## -> x when completing
}
} zsh -f
} touch xenon
} autoload compinit; compinit
} setopt globcomplete
} cat \##<tab> -> cat xenon

Hmm.  This is probably happening because the completion internals use
an "x" as a place-holder in some instances.  See the functions
comptils.c:comp_quote, compcore.c:tildequote, and zle_tricky.c:addx,
the latter of which has a big comment explaining what's going on.

However, I think it's more likely compcore.c:set_comp_sep that is the
culprit.  We can find out by changing the 'x' to something else in one
of addx or set_comp_sep and then trying the completion again.

Which reveals that it's neither of those; rather compcore.c:addmatches
is playing funny with us, at line 2302:

2293         if (comppatmatch && *comppatmatch) {
2294             int is = (*comppatmatch == '*');
2295             char *tmp = (char *) zhalloc(2 + llpl + llsl + gfl);
2296 
2297             if (gfl) {
2298                 strcpy(tmp, globflag);
2299                 strcat(tmp, lpre);
2300             } else
2301                 strcpy(tmp, lpre);
2302             tmp[llpl + gfl] = 'x';
2303             strcpy(tmp + llpl + gfl + is, lsuf);
2304 
2305             tokenize(tmp);
2306             remnulargs(tmp);
2307             if (haswilds(tmp)) {
2308                 if (is)
2309                     tmp[llpl + gfl] = Star;
2310                 if ((cp = patcompile(tmp, 0, NULL)))
2311                     haspattern = 1;
2312             }
2313         }

The problem seems to be that remnulargs() has shortened tmp so that
the count (llpl + gfl) is no longer correct.  I believe we can fix
that by moving remnulargs to after line 2309.

Index: Src/Zle/compcore.c
===================================================================
--- Src/Zle/compcore.c	20 Dec 2011 17:13:38 -0000	1.29
+++ Src/Zle/compcore.c	9 Feb 2012 15:33:44 -0000
@@ -2303,10 +2303,10 @@
 		strcpy(tmp + llpl + gfl + is, lsuf);
 
 		tokenize(tmp);
-		remnulargs(tmp);
 		if (haswilds(tmp)) {
 		    if (is)
 			tmp[llpl + gfl] = Star;
+		    remnulargs(tmp);
 		    if ((cp = patcompile(tmp, 0, NULL)))
 			haspattern = 1;
 		}

Re: \## -> x when completing

[Mikael Magnusson]

On 9 February 2012 16:37, Bart Schaefer <schaefer@brasslantern.com> wrote:
> This gets a bit stream-of-consciousness but maybe that's helpful.
>
> On Feb 8, 11:37pm, Mikael Magnusson wrote:
> } Subject: \## -> x when completing
> }
> } zsh -f
> } touch xenon
> } autoload compinit; compinit
> } setopt globcomplete
> } cat \##<tab> -> cat xenon
>
> Hmm.  This is probably happening because the completion internals use
> an "x" as a place-holder in some instances.  See the functions
> comptils.c:comp_quote, compcore.c:tildequote, and zle_tricky.c:addx,
> the latter of which has a big comment explaining what's going on.
>
> However, I think it's more likely compcore.c:set_comp_sep that is the
> culprit.  We can find out by changing the 'x' to something else in one
> of addx or set_comp_sep and then trying the completion again.
>
> Which reveals that it's neither of those; rather compcore.c:addmatches
> is playing funny with us, at line 2302:
>
> 2293         if (comppatmatch && *comppatmatch) {
> 2294             int is = (*comppatmatch == '*');
> 2295             char *tmp = (char *) zhalloc(2 + llpl + llsl + gfl);
> 2296
> 2297             if (gfl) {
> 2298                 strcpy(tmp, globflag);
> 2299                 strcat(tmp, lpre);
> 2300             } else
> 2301                 strcpy(tmp, lpre);
> 2302             tmp[llpl + gfl] = 'x';
> 2303             strcpy(tmp + llpl + gfl + is, lsuf);
> 2304
> 2305             tokenize(tmp);
> 2306             remnulargs(tmp);
> 2307             if (haswilds(tmp)) {
> 2308                 if (is)
> 2309                     tmp[llpl + gfl] = Star;
> 2310                 if ((cp = patcompile(tmp, 0, NULL)))
> 2311                     haspattern = 1;
> 2312             }
> 2313         }
>
> The problem seems to be that remnulargs() has shortened tmp so that
> the count (llpl + gfl) is no longer correct.  I believe we can fix
> that by moving remnulargs to after line 2309.
>
> Index: Src/Zle/compcore.c
> ===================================================================
> --- Src/Zle/compcore.c  20 Dec 2011 17:13:38 -0000      1.29
> +++ Src/Zle/compcore.c  9 Feb 2012 15:33:44 -0000
> @@ -2303,10 +2303,10 @@
>                strcpy(tmp + llpl + gfl + is, lsuf);
>
>                tokenize(tmp);
> -               remnulargs(tmp);
>                if (haswilds(tmp)) {
>                    if (is)
>                        tmp[llpl + gfl] = Star;
> +                   remnulargs(tmp);
>                    if ((cp = patcompile(tmp, 0, NULL)))
>                        haspattern = 1;
>                }

This fix seems to work for me, I have barely tried anything but my
test case so far though :). cat \##g<tab> now works in an irclogs dir
to complete both files starting with g and #g though.

-- 
Mikael Magnusson

Re: \## -> x when completing

[Bart Schaefer]

On Thu, Feb 9, 2012 at 8:16 AM, Mikael Magnusson <mikachu@gmail.com> wrote:
>>                tokenize(tmp);
>> -               remnulargs(tmp);
>>                if (haswilds(tmp)) {
>>                    if (is)
>>                        tmp[llpl + gfl] = Star;
>> +                   remnulargs(tmp);
>>                    if ((cp = patcompile(tmp, 0, NULL)))
>>                        haspattern = 1;
>>                }
>
> This fix seems to work for me, I have barely tried anything but my
> test case so far though :). cat \##g<tab> now works in an irclogs dir

My new worry is that tokenize() might make the string *longer*, thus
causing the Star to be inserted too far to the left.