Boundary error in VCS_INFO_set --nvcs

Boundary error in VCS_INFO_set --nvcs

[Bart Schaefer]

This output is from an older version of VCS_INFO but I don't see a
diff that would have changed anything but line numbers:

+vcs_info:52> VCS_INFO_set --nvcs
+VCS_INFO_set:5> setopt localoptions noksharrays NO_shwordsplit
+VCS_INFO_set:6> local -i i j
+VCS_INFO_set:8> [[ --nvcs == --nvcs ]]
+VCS_INFO_set:9> [[ '' == -preinit- ]]
+VCS_INFO_set:10> i=0
+VCS_INFO_set:11> typeset -gx 'vcs_info_msg_0_='
+VCS_INFO_set:10> i=-1
+VCS_INFO_set:11> typeset -gx 'vcs_info_msg_-1_='
VCS_INFO_set:typeset:11: not valid in this context: vcs_info_msg_-1_

The problem is that when disable-patterns matches the current
directory, vcs_info invokes VCS_INFO_set --nvcs, which hits this loop:

 if [[ $1 == '--nvcs' ]] ; then
     [[ $2 == '-preinit-' ]] && (( maxexports == 0 )) && (( maxexports = 1 ))
     for i in {0..$((maxexports - 1))} ; do

Because $2 is not "-preeinit-", when maxexports is zero it is never
reset to 1 and the "for" loop runs backwards to -1.  I don't otherwise
know how to reproduce the situation where VCS_INFO_maxexports is never
called.

Re: Boundary error in VCS_INFO_set --nvcs

[Daniel Shahaf]

Bart Schaefer wrote on Wed, Apr 13, 2022 at 14:34:28 -0700:
> This output is from an older version of VCS_INFO but I don't see a
> diff that would have changed anything but line numbers:
> 
> +vcs_info:52> VCS_INFO_set --nvcs
> +VCS_INFO_set:5> setopt localoptions noksharrays NO_shwordsplit
> +VCS_INFO_set:6> local -i i j
> +VCS_INFO_set:8> [[ --nvcs == --nvcs ]]
> +VCS_INFO_set:9> [[ '' == -preinit- ]]
> +VCS_INFO_set:10> i=0
> +VCS_INFO_set:11> typeset -gx 'vcs_info_msg_0_='
> +VCS_INFO_set:10> i=-1
> +VCS_INFO_set:11> typeset -gx 'vcs_info_msg_-1_='
> VCS_INFO_set:typeset:11: not valid in this context: vcs_info_msg_-1_
> 
> The problem is that when disable-patterns matches the current
> directory, vcs_info invokes VCS_INFO_set --nvcs, which hits this loop:
> 
>  if [[ $1 == '--nvcs' ]] ; then
>      [[ $2 == '-preinit-' ]] && (( maxexports == 0 )) && (( maxexports = 1 ))
>      for i in {0..$((maxexports - 1))} ; do
> 
> Because $2 is not "-preeinit-", when maxexports is zero it is never
> reset to 1 and the "for" loop runs backwards to -1.  I don't otherwise
> know how to reproduce the situation where VCS_INFO_maxexports is never
> called.

Can't reproduce.  What's $ZSH_PATCHLEVEL?  Could you try in 'zsh -f' in
a new directory (create a new repository therein)?

Cheers,

Daniel

Re: Boundary error in VCS_INFO_set --nvcs

[Daniel Shahaf]

Daniel Shahaf wrote on Sun, Apr 17, 2022 at 18:04:23 +0000:
> Bart Schaefer wrote on Wed, Apr 13, 2022 at 14:34:28 -0700:
> > This output is from an older version of VCS_INFO but I don't see a
> > diff that would have changed anything but line numbers:
> > 
> > +vcs_info:52> VCS_INFO_set --nvcs
> > +VCS_INFO_set:5> setopt localoptions noksharrays NO_shwordsplit
> > +VCS_INFO_set:6> local -i i j
> > +VCS_INFO_set:8> [[ --nvcs == --nvcs ]]
> > +VCS_INFO_set:9> [[ '' == -preinit- ]]
> > +VCS_INFO_set:10> i=0
> > +VCS_INFO_set:11> typeset -gx 'vcs_info_msg_0_='
> > +VCS_INFO_set:10> i=-1
> > +VCS_INFO_set:11> typeset -gx 'vcs_info_msg_-1_='
> > VCS_INFO_set:typeset:11: not valid in this context: vcs_info_msg_-1_
> > 
> > The problem is that when disable-patterns matches the current
> > directory, vcs_info invokes VCS_INFO_set --nvcs, which hits this loop:
> > 
> >  if [[ $1 == '--nvcs' ]] ; then
> >      [[ $2 == '-preinit-' ]] && (( maxexports == 0 )) && (( maxexports = 1 ))
> >      for i in {0..$((maxexports - 1))} ; do
> > 
> > Because $2 is not "-preeinit-", when maxexports is zero it is never
> > reset to 1 and the "for" loop runs backwards to -1.  I don't otherwise
> > know how to reproduce the situation where VCS_INFO_maxexports is never
> > called.
> 
> Can't reproduce.  What's $ZSH_PATCHLEVEL?  Could you try in 'zsh -f' in
> a new directory (create a new repository therein)?

Found a(nother?) way.  In current HEAD and 'zsh -f', call vcs_info once,
_then_ set «zstyle ':vcs_info:*' enable none», and the error will happen
via the following callsite:

    96	    if [[ -n ${(M)enabled:#(#i)none} ]] ; then
    97	        [[ -n ${vcs_info_msg_0_} ]] && VCS_INFO_set --nvcs
    98	        return 0
    99	    fi

The documentation of the «nvcsformats» style makes it clear that in the
"no VCS" case vcs_info_msg_0_ should still get set.

I'm not sure, however, what the right code change is.  Anyone wants to
jump in?

This is not a regression, and as such not a blocker.

Cheers,

Daniel

Re: Boundary error in VCS_INFO_set --nvcs

[Bart Schaefer]

On Sun, Apr 17, 2022 at 11:48 AM Daniel Shahaf <d.s@daniel.shahaf.name> wrote:
>
> > Bart Schaefer wrote on Wed, Apr 13, 2022 at 14:34:28 -0700:
> > >      [[ $2 == '-preinit-' ]] && (( maxexports == 0 )) && (( maxexports = 1 ))
> > >      for i in {0..$((maxexports - 1))} ; do
>
> I'm not sure, however, what the right code change is.  Anyone wants to
> jump in?

Dropping [[ $2 == '-preinit-' ]] && would be the obvious thing, but
I'm not sure why we care about it, so perhaps something more
complicated is required.