Re: [mpol@charybda.icm.edu.pl: Bug#306346: zsh: replacement seems broken in prompt]

[Bart Schaefer <schaefer@brasslantern.com>]

On Apr 25,  9:37pm, Clint Adams wrote:
} Subject: [mpol@charybda.icm.edu.pl: Bug#306346: zsh: replacement seems bro
}
} This is with 21170.
} 
} ----- Forwarded message from Michal Politowski <mpol@charybda.icm.edu.pl> -----
} 
} ${${foo}/?*/replacement} puts replacement in the prompt even when foo is empty

Hrm.  This seems to be strictly a promptsubst thing, because

schaefer<505> echo ${${foo}/?*/replacement}

schaefer<506> RPS1='${${foo}/?*/replacement}'
schaefer<507> setopt promptsubst                       ${${foo}/?*/replacement}
schaefer<508>                                                       replacement

Ah, here's an easier way to reproduce it:

schaefer<510> echo "${${foo}/?*/replacement}"
replacement

So it's a quoting thing.

According to GDB, both before and after 21170 the inner ${foo} is
resulting in "\233\0" (which is a Nularg).  After 21170, however, the
comparison against ?* matches this as one character.

The problem is here at line 2344:

2344                        if (pattrylen(p, t, s + l - t, umlen, ioff)) {

And it seems that this is the answer to the question in pattern.c:

    /* inherited from domatch, but why, exactly? */
    if (*string == Nularg)
	string++;

This is precisely the case we're encountering here; it changes the value
of unmetalen within pattryrefs() in a way that we're not changing it
when precomputing it in igetmatch().

What I don't know, though, is whether the Nularg test should be copied
into both igetmatch() and down to by the ztrsub() call in pattryrefs(),
or whether it should simply be

    /* inherited from domatch, but why, exactly? */
    if (*string == Nularg)
	string++, unmetalen--;