Re: Latest CVS: configure --pre=/to does not work

Re: Latest CVS: configure --pre=/to does not work

[Sven Wischnowsky]

Andrej Borsenkow wrote:

> Just after CVS update about 15 min before ... (18:00 Moscow time - GMT+3)
> 
> Why every completion update breakes something ... :-)

Because it *is* quite complicated, so much so that I can't tests
everything.

Sorry.

> bor@itsrm2:~/zsh%> zsh -f
> itsrm2% autoload -U compinit; compinit -D 
> itsrm2% ./configure --pre=/to<TAB> 
> itsrm2% ./configure --pre=/toopt=''
> <== Cursor here

The `opt=...' is the output of a `local' that got a parameter that ws
already made local by a previous `local'.

The missing completions were caused by a wrong test in `_arguments'
(and a missing `%%:*').

The patch also ensures that even if multiple calls to `comptry' add
the same tag, this tag will be used in only the first of theses
tags. For the functions we currently have, this is the right thing
(otherwise we might execute code more than once where the first
execution showed that it's useless to try this again). However, one
might argue that under certain circumstances it would be nice to allow 
tags to be in more than one set (if tags were used slightly different
from the way they are used now). If we ever find a case where this is
useful, I'll add an option to either `_tags' or `comptry' to allow
that.


This patch also fixes this for me:

Tanaka Akira wrote:

> Z:akr@is27e1u11% Src/zsh -f
> is27e1u11% bindkey -e; fpath=($PWD/Completion/*(/)); autoload -U compinit; compinit -D; compdef _tst tst
> is27e1u11% cvs <TAB>
> CVS/           Src/           commit        init          rtag
> Completion/    StartupFiles/  diff          log           status
> Config/        Util/          edit          login         tag
> Doc/           add           editors       logout        unedit
> Etc/           admin         export        rdiff         update
> Functions/     annotate      history       release       watch
> Misc/          checkout      import        remove        watchers
> 
> Wow. filenames shouldn't listed.
> 
> And after second <TAB>:
> 
> is27e1u11% cvs --allow-root=CVS/
> 
> Of cause, the option `--allow-root' shouldn't completed.

That was caused by the wrong test, too.

Bye
 Sven

diff -u oldcompletion/Base/_arguments Completion/Base/_arguments
--- oldcompletion/Base/_arguments	Fri Nov 19 09:55:53 1999
+++ Completion/Base/_arguments	Mon Nov 22 10:16:47 1999
@@ -24,7 +24,7 @@
   name="${name//[^a-zA-Z0-9_]/_}"
 
   if (( ! ${(P)+name} )); then
-    local iopts sopts pattern tmpo cur opt cache
+    local iopts sopts pattern tmpo cur cache
     typeset -U lopts
 
     cache=()
@@ -287,8 +287,7 @@
         fi
       fi
       if [[ -n "$opts" && -z "$aret$matched" &&
-            nm -ne compstate[nmatches] ]] &&
-          _requested arguments; then
+            nm -eq compstate[nmatches] ]]; then
 
         local prefix suffix
 
@@ -303,8 +302,10 @@
 	  SUFFIX="$suffix"
 	  IPREFIX="${IPREFIX}${equal[1]%%:*}="
 	  matched=yes
-	  comparguments -L "$equal[1]" descr action subc
+
+	  comparguments -L "${equal[1]%%:*}" descr action subc
 	  curcontext="${oldcontext}:$subc"
+
 	  continue
         fi
       fi
diff -u oldsrc/Zle/computil.c Src/Zle/computil.c
--- oldsrc/Zle/computil.c	Tue Nov 16 10:07:52 1999
+++ Src/Zle/computil.c	Mon Nov 22 10:19:24 1999
@@ -2730,8 +2730,16 @@
 	args = arrdup(args);
 
 	for (p = q = args, all = comptags[lasttaglevel]->all; *p; p++)
-	    if (arrcontains(all, *p))
-		*q++ = *p;
+	    if (arrcontains(all, *p)) {
+		Ctset s;
+
+		for (s = comptags[lasttaglevel]->sets; s; s = s->next)
+		    if (arrcontains(s->tags, *p))
+			break;
+
+		if (!s)
+		    *q++ = *p;
+	    }
 	*q = NULL;
 
 	if (*args) {

--
Sven Wischnowsky                         wischnow@informatik.hu-berlin.de

Completion test suite

[Tanaka Akira]

In article <199911220934.KAA04967@beta.informatik.hu-berlin.de>,
  Sven Wischnowsky <wischnow@informatik.hu-berlin.de> writes:

> Because it *is* quite complicated, so much so that I can't tests
> everything.

So, definitely we need some automatic test suite for completion
system.

When I tried it with expect, I found that it is very difficult because
expect's support of screen based user interfaces like zsh (and
other termcap/terminfo based program and curses based program) is not
so good.

I think that zsh should have some noninteractive way to use completion
system for this purpose.  If we can invoke completion system
noninteractively with an arbitrary command line and zsh can dumps
its completion result, it is enough to write test suite, maybe.
-- 
Tanaka Akira

Re: Latest CVS: configure --pre=/to does not work

[Bart Schaefer]

(Responding to 8716, which I pulled from my archive account as I mentioned.)

On Nov 22, 11:27am, Sven Wischnowsky wrote:
} Subject: Re: Latest CVS: configure --pre=/to does not work
}
} This patch also fixes this for me:
} 
} Tanaka Akira wrote:
} 
} > Z:akr@is27e1u11% Src/zsh -f
} > is27e1u11% bindkey -e; fpath=($PWD/Completion/*(/)); autoload -U compinit; compinit -D; compdef _tst tst
} > is27e1u11% cvs <TAB>
} > CVS/           Src/           commit        init          rtag
} > Completion/    StartupFiles/  diff          log           status
} > Config/        Util/          edit          login         tag
} > Doc/           add           editors       logout        unedit
} > Etc/           admin         export        rdiff         update
} > Functions/     annotate      history       release       watch
} > Misc/          checkout      import        remove        watchers
} > 
} > Wow. filenames shouldn't listed.
} > 
} > And after second <TAB>:
} > 
} > is27e1u11% cvs --allow-root=CVS/
} > 
} > Of cause, the option `--allow-root' shouldn't completed.
} 
} That was caused by the wrong test, too.

I now get the following equally strange behavior:

zagzig<3> cvs <TAB>
add        commit     export     log        release    tag        watchers 
admin      diff       history    login      remove     unedit              
annotate   edit       import     logout     rtag       update              
checkout   editors    init       rdiff      status     watch               
zagzig<3> cvs -<TAB>
(FEEP)
zagzig<3> cvs - <TAB>
zagzig<3> cvs - --allow-root=

And --allow-root= is still the *only* option it will complete, even though
several others are passed to "_arguments -s" by _cvs.

-- 
Bart Schaefer                                 Brass Lantern Enterprises
http://www.well.com/user/barts              http://www.brasslantern.com

Re: Completion test suite

[Sven Wischnowsky]

Tanaka Akira wrote:

> In article <199911220934.KAA04967@beta.informatik.hu-berlin.de>,
>   Sven Wischnowsky <wischnow@informatik.hu-berlin.de> writes:
> 
> > Because it *is* quite complicated, so much so that I can't tests
> > everything.
> 
> So, definitely we need some automatic test suite for completion
> system.
> 
> When I tried it with expect, I found that it is very difficult because
> expect's support of screen based user interfaces like zsh (and
> other termcap/terminfo based program and curses based program) is not
> so good.
> 
> I think that zsh should have some noninteractive way to use completion
> system for this purpose.  If we can invoke completion system
> noninteractively with an arbitrary command line and zsh can dumps
> its completion result, it is enough to write test suite, maybe.

I don't think so. First, this would leave the problem unsolved that
we have to find out that the results or ok. I.e. either some pour
sould would have to go through the output and look if it's correct or
everyone who writes a completion function would have to extend the
test script so that it can check it automatically. Considering the
amount that would have to be tested (and all the dependencies on the
environment: files that have to existent, different settings for
styles, different _sort_tags functions, commands that have to be
existent in some version or not, compmatchers,...) noone will
volunteer for the first possibility. And since most people are even to 
lazy to write descriptions I'm sure that most people would be to lazy
to add test-code. And that would have to be correct under all
circumstances, too. Sounds completely impossible to me.

The other problem is that the completion code relies in some places on 
zle being active. In some cases it is even visible: the special zle
parameters that are used by some completion functions. In other cases
it is hidden, but, for example, the listing is really done by zle
(complete supplies the function but that is invoked by zle). Or think
about the auto_menu stuff, where we rely on zle to set some variables.

Etc, etc...

In short: I don't think I would even try to write that, but I would be 
willing to give the implementor every help I can, of course.

Bye
 Sven

P.S.: Yes, sure, we could at least try to test some of the things
      completion does: is the correct string inserted and are the
      correct matches being generated. But that would still leave us
      with the problems of the first form.

--
Sven Wischnowsky                         wischnow@informatik.hu-berlin.de