Re: configure --with- completion

Re: configure --with- completion

[Vincent Lefevre]

[Moved to workers]

On 2015-09-04 10:40:27 +0200, Vincent Lefevre wrote:
> In MPFR, I can do:
> 
>   ./configure --with-gmp=[TAB]
>   ./configure --with-gmp-include=[TAB]
>   ./configure --with-gmp-lib=[TAB]
> 
> and complete with directory names, but
> 
>   ./configure --with-gmp-build=[TAB]
> 
> doesn't allow me to complete with directory names.
> 
> FYI, "./configure --with-[TAB]" outputs:
> 
> Completing option
> --with-gmp              -- GMP install directory
> --with-gmp-build        -- GMP build directory (please read INSTALL file)
> --with-gmp-include      -- GMP include directory
> --with-gmp-lib          -- GMP lib directory
> --with-gnu-ld           -- assume the C compiler uses GNU ld (default=no)
> --with-mulhigh-size     -- internal threshold table for mulhigh
> --with-pic              -- try to use only PIC/non-PIC objects (default=use
> --with-sysroot          -- Search for dependent libraries within DIR
> --without-gmp           -- GMP install directory
> --without-gmp-build     -- GMP build directory (please read INSTALL file)
> --without-gmp-include   -- GMP include directory
> --without-gmp-lib       -- GMP lib directory
> --without-gnu-ld        -- assume the C compiler uses GNU ld (default=no)
> --without-mulhigh-size  -- internal threshold table for mulhigh
> --without-pic           -- try to use only PIC/non-PIC objects (default=use
> --without-sysroot       -- Search for dependent libraries within DIR
> 
> and "./configure --help" outputs:
> 
> [...]
>   --with-gmp-include=DIR  GMP include directory
>   --with-gmp-lib=DIR      GMP lib directory
>   --with-gmp=DIR          GMP install directory
>   --with-gmp-build=DIR    GMP build directory (please read INSTALL file)
> [...]
> 
> Is there something missing in zsh completion?

The problem seems to come to the fact that

  --with-gmp-build=DIR    GMP build directory (please read INSTALL file)

is listed after

  --with-gmp=DIR          GMP install directory

by "./configure --help". If I switch these lines, everything works
as expected. There's no reason to depend on the order of the options
listed by "./configure --help". So, this seems to be a bug somewhere
in the zsh completion system.

-- 
Vincent Lefèvre <vincent@vinc17.net> - Web: <https://www.vinc17.net/>
100% accessible validated (X)HTML - Blog: <https://www.vinc17.net/blog/>
Work: CR INRIA - computer arithmetic / AriC project (LIP, ENS-Lyon)

Re: configure --with- completion

[Jun T.]

2015/10/20 22:00, Vincent Lefevre <vincent@vinc17.net> wrote:

> The problem seems to come to the fact that
> 
>  --with-gmp-build=DIR    GMP build directory (please read INSTALL file)
> 
> is listed after
> 
>  --with-gmp=DIR          GMP install directory

Yes. For example, 

% _foo() { 
_arguments : '-x=:switch:(on off)' '-xy=:answer:(yes no)' '*:file:_files'
}
% compdef _foo foo
% foo -xy=<TAB>

this completes nothing. If -x and -xy are exchanged, it works as expected.

In ca_get_opt() (computil.c:1167), the current word seems to be compared
with optspecs in the order they are passed to _arguments.  In the example
above, the current word "-xy=" is first compared with '-x', and since
'-x' matches with the beginning of "-xy=" (strpfx() at line 1695),
it is considered as the option -x followed by 'y='. So the next (correct)
optspec '-xy' is not tried.

Probably we need to look at the next character after the option '-x'
('y' in the above example), and if the type of the option is CAO_OEQUAL
and the next character is not '=' (or NULL), then reject '-x' and 
go to the next possible option '-xy'. But I rather hesitate to modify
functions which I don't understand well yet.

A workaround (for '_arguments --') would be to modify _arguments as in
the patch below. It sorts the optspecs generated by '_arguments --'
so that --with-gmp-build comes before --with-gmp.

But this patch has effect only with '_arguments --', and does not fix
the example above (-x and -xy). Patching computil.c would be better ...



diff --git a/Completion/Base/Utility/_arguments b/Completion/Base/Utility/_arguments
index 87fb20e..fa9750e 100644
--- a/Completion/Base/Utility/_arguments
+++ b/Completion/Base/Utility/_arguments
@@ -303,7 +303,13 @@ if (( long )); then
         fi
       fi
     done
-    set -A "$name" "${(@)cache:# #}"
+    # Sort the optspecs in $cache so that -xy comes before -x (this
+    # seems to be required by a restriction (or bug) of comparguments).
+    # We do this by the following steps:
+    #     add '~' at the end of option name,  sort,  and remove '~'.
+    #     ('~'=0x7e comes after any other printable ASCII characters)
+    tmp=( ${${(o)${${cache:# #}/(#b)([a-zA-Z0-9_-]#)/$match[1]~}}/\~/} )
+    set -A "$name" "$tmp[@]"
   fi
   set -- "$tmpargv[@]" "${(@P)name}"
 fi

Re: configure --with- completion

[Jun T.]

On 2015/10/22, at 1:11, Jun T. <takimoto-j@kba.biglobe.ne.jp> wrote:
> 
> But this patch has effect only with '_arguments --', and does not fix
> the example above (-x and -xy). Patching computil.c would be better ...

So this is a minimal patch I can think of.
Is this reasonable?

diff --git a/Src/Zle/computil.c b/Src/Zle/computil.c
index e5db086..b664f30 100644
--- a/Src/Zle/computil.c
+++ b/Src/Zle/computil.c
@@ -1693,10 +1693,13 @@ ca_get_opt(Cadef d, char *line, int full, char **end)
 	for (p = d->opts; p; p = p->next)
 	    if (p->active && ((!p->args || p->type == CAO_NEXT) ?
 			      !strcmp(p->name, line) : strpfx(p->name, line))) {
+		int l = strlen(p->name);
+		if ((p->type == CAO_OEQUAL || p->type == CAO_EQUAL) &&
+		    line[l] && line[l] != '=')
+		    continue;
+
 		if (end) {
 		    /* Return a pointer to the end of the option. */
-		    int l = strlen(p->name);
-
 		    if ((p->type == CAO_OEQUAL || p->type == CAO_EQUAL) &&
 			line[l] == '=')
 			l++;

Re: configure --with- completion

[Peter Stephenson]

On Thu, 22 Oct 2015 16:51:54 +0900
"Jun T." <takimoto-j@kba.biglobe.ne.jp> wrote:
> On 2015/10/22, at 1:11, Jun T. <takimoto-j@kba.biglobe.ne.jp> wrote:
> > 
> > But this patch has effect only with '_arguments --', and does not fix
> > the example above (-x and -xy). Patching computil.c would be better ...
> 
> So this is a minimal patch I can think of.
> Is this reasonable?

In this part of the code, committing it and seeing if it has any
negative effects is about the only way of finding out.

pws