[PATCH] completion: matcher correspondence classes: fix offset

[PATCH] completion: matcher correspondence classes: fix offset

[m0viefreak]

The "compfiles" function passes the matcher through the
"cfp_matcher_range" function to turn a prefix into
character classes. Example:

  matcher: m:{a-z}={A-Z}
  prefix: fo<tab>
  file list: foo1.txt FOO2.txt

the cfp_matcher_range function is supposed to turn the
prefix into:

  [fF][oO]

However there is bug which causes the offset to shift
by 1. Without this patch it returns:

  [fE][oN]

The bug was unnoticed before, because in the usual
case -- that is when matcher-list is set -- _path_files
calls compfiles -p ... "m:{a-z}={A-Z} m:{a-z}={A-Z}"
with the matcher added twice. This causes
"cfp_matcher_range" to do nothing at all, and instead
the whole list of files in the directory is returned.
Then it is filtered using compadd -D.

A case where this bug surfaces is the following:

Make sure no global matcher-list is set and set a single
matcher instead:

  zstyle -d ':completion:*' matcher-list
  zstyle ':completion:*' matcher 'm:{a-z}={A-Z}'

Now the completion

  cat fo<tab>

only offers "foo1.txt" but not "FOO2.txt".

Questions I couldn't anwswer, yet:
a) Why does "cfp_matcher_range" do nothing if more than
   one matcher spec is given?
b) Do we even need the "cfp_matcher_range" function
   at all? Seems like the later compadd -D works just
   fine and we could remove a whole bunch of unneeded
   functions that are only called from cfp_matcher_range.
c) Why does _path_files duplicate all matcher-list
   entries when calling "compfiles -p"?
---
 Src/Zle/compmatch.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/Src/Zle/compmatch.c b/Src/Zle/compmatch.c
index 0e41ac3..667d71d 100644
--- a/Src/Zle/compmatch.c
+++ b/Src/Zle/compmatch.c
@@ -1218,7 +1218,7 @@ pattern_match_equivalence(Cpattern lp, convchar_t wind, int wmtp,
     convchar_t lchr;
     int lmtp;
 
-    if (!PATMATCHINDEX(lp->u.str, wind-1, &lchr, &lmtp)) {
+    if (!PATMATCHINDEX(lp->u.str, wind, &lchr, &lmtp)) {
 	/*
 	 * No equivalent.  No possible match; give up.
 	 */
-- 
2.5.0.234.gefc8a62

Re: [PATCH] completion: matcher correspondence classes: fix offset

[Mikael Magnusson]

Were there ever any follwoups to this? I don't see any in the archive,
and the code still has the -1.

On 3/13/16, m0viefreak <m0viefreak.cm@googlemail.com> wrote:
> The "compfiles" function passes the matcher through the
> "cfp_matcher_range" function to turn a prefix into
> character classes. Example:
>
>   matcher: m:{a-z}={A-Z}
>   prefix: fo<tab>
>   file list: foo1.txt FOO2.txt
>
> the cfp_matcher_range function is supposed to turn the
> prefix into:
>
>   [fF][oO]
>
> However there is bug which causes the offset to shift
> by 1. Without this patch it returns:
>
>   [fE][oN]
>
> The bug was unnoticed before, because in the usual
> case -- that is when matcher-list is set -- _path_files
> calls compfiles -p ... "m:{a-z}={A-Z} m:{a-z}={A-Z}"
> with the matcher added twice. This causes
> "cfp_matcher_range" to do nothing at all, and instead
> the whole list of files in the directory is returned.
> Then it is filtered using compadd -D.
>
> A case where this bug surfaces is the following:
>
> Make sure no global matcher-list is set and set a single
> matcher instead:
>
>   zstyle -d ':completion:*' matcher-list
>   zstyle ':completion:*' matcher 'm:{a-z}={A-Z}'
>
> Now the completion
>
>   cat fo<tab>
>
> only offers "foo1.txt" but not "FOO2.txt".
>
> Questions I couldn't anwswer, yet:
> a) Why does "cfp_matcher_range" do nothing if more than
>    one matcher spec is given?
> b) Do we even need the "cfp_matcher_range" function
>    at all? Seems like the later compadd -D works just
>    fine and we could remove a whole bunch of unneeded
>    functions that are only called from cfp_matcher_range.
> c) Why does _path_files duplicate all matcher-list
>    entries when calling "compfiles -p"?
> ---
>  Src/Zle/compmatch.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/Src/Zle/compmatch.c b/Src/Zle/compmatch.c
> index 0e41ac3..667d71d 100644
> --- a/Src/Zle/compmatch.c
> +++ b/Src/Zle/compmatch.c
> @@ -1218,7 +1218,7 @@ pattern_match_equivalence(Cpattern lp, convchar_t
> wind, int wmtp,
>      convchar_t lchr;
>      int lmtp;
>
> -    if (!PATMATCHINDEX(lp->u.str, wind-1, &lchr, &lmtp)) {
> +    if (!PATMATCHINDEX(lp->u.str, wind, &lchr, &lmtp)) {
>  	/*
>  	 * No equivalent.  No possible match; give up.
>  	 */
> --
> 2.5.0.234.gefc8a62
>
>


-- 
Mikael Magnusson

Re: [PATCH] completion: matcher correspondence classes: fix offset

[Bart Schaefer]

On Thu, Dec 23, 2021 at 12:25 PM Mikael Magnusson <mikachu@gmail.com> wrote:
>
> Were there ever any follwoups to this?

I don't recall seeing any.  Probably the only thing desirable beyond
that patch would be a regression test.

In the original message:
> > a) Why does "cfp_matcher_range" do nothing if more than
> >   one matcher spec is given?

It's not that it does nothing, it's that it never gets called by
cfp_matcher_pats() when there is a matcher-list style but not a
matcher style.  I haven't worked through exactly why THAT happens, but
it has to do with whether there's anything left over after the loop in
cfp_matcher_pats().  I have noticed that the "compadd -D" pass behaves
the same way regardless of the styles:  cfp_matcher_pats() loops
through the matcher spec and ends up with nothing left at the end, so
cfp_matcher_range() is not called.

> > b) Do we even need the "cfp_matcher_range" function
> >   at all? Seems like the later compadd -D works

Although _path_files is the only function that currently calls
compfiles -p, that doesn't mean no other function ever might ... so
what would happen if the compadd -D were never done?

> > c) Why does _path_files duplicate all matcher-list
> >   entries when calling "compfiles -p"?

It doesn't appear to duplicate matcher-list entries ... what it does
is build its own list of matchers from its -M option, the "expl" value
returned by _description, and the setting of nocaseglob, and then
appends that to the matcher-list passed down from _main_complete.
That may or may not duplicate parts of the matcher-list.  I suppose
_path_files could scan the $_matcher global to be sure it doesn't
append something that is already there ...?

Re: [PATCH] completion: matcher correspondence classes: fix offset

[Bart Schaefer]

On Fri, Dec 24, 2021 at 11:58 AM Bart Schaefer
<schaefer@brasslantern.com> wrote:
>
> Probably the only thing desirable beyond
> that patch would be a regression test.

I've gone ahead and commited the patch.