_make suppress error on OpenBSD.

_make suppress error on OpenBSD.

[Tanaka Akira]

Z:akr@coulee% Src/zsh -f
coulee% bindkey -e; autoload -U compinit; compinit -D
coulee% make awk: i/o error occurred while closing /dev/stdin
 input record number 366, file Makefile
 source line number 1

FORCE                   config                  install.bin             mostlyclean-recursive   test
META-FAQ                config.h                install.fns             prep                    uninstall
Makefile                config.status           install.html            realclean               uninstall.bin
all                     distclean               install.info            realclean-here          uninstall.fns
check                   distclean-here          install.man             realclean-recursive     uninstall.html
clean                   distclean-recursive     install.modules         stamp-h                 uninstall.info
clean-here              install                 mostlyclean             targz-doc               uninstall.man
clean-recursive         install-strip           mostlyclean-here        targz-src               uninstall.modules

This patch suppress this error.

Index: Completion/User/_make
===================================================================
RCS file: /cvsroot/zsh/zsh/Completion/User/_make,v
retrieving revision 1.7
diff -u -r1.7 _make
--- Completion/User/_make	2000/04/27 12:44:21	1.7
+++ Completion/User/_make	2000/05/09 12:04:50
@@ -39,7 +39,7 @@
             $(awk '/^[a-zA-Z0-9][^\/\t=]+:/ {print $1}
  	      /^\.include  *<bsd\.port\.(subdir\.|pre\.)?mk>/ || /^\.include  *".*mk\/bsd\.pkg\.(subdir\.)?mk"/ {
  	        print "fetch fetch-list extract patch configure build install reinstall deinstall package describe checkpatch checksum makesum" }' \
- 	     FS=: $file)
+ 	     FS=: $file </dev/null)
            )
     fi
     _wanted targets expl 'make target' compadd "$tmp[@]" && return 0
-- 
Tanaka Akira

closing stdin (Re: _make suppress error on OpenBSD.)

[Tanaka Akira]

In article <hvoog6ggc41.fsf@serein.m17n.org>,
  Tanaka Akira <akr@m17n.org> writes:

> Z:akr@coulee% Src/zsh -f
> coulee% bindkey -e; autoload -U compinit; compinit -D
> coulee% make awk: i/o error occurred while closing /dev/stdin
>  input record number 366, file Makefile
>  source line number 1

I found same problem in zfcd_match.  It can easily fixed but I like
more fundamental fix.

This problem is caused by the one true awk behavior.  (OpenBSD comes
with the one true awk.)  The one true awk closes all file descriptors
which should be opened when exit and complain if it causes an error.
Of course, it assumes that stdin is opened.

http://cm.bell-labs.com/who/bwk/awk.tar.gz

However, we can suppress the error by modifying each awk invocation,
but I think that completion system should redirect stdin from
/dev/null instead of closing it.
-- 
Tanaka Akira

Re: closing stdin (Re: _make suppress error on OpenBSD.)

[Bart Schaefer]

On May 10,  4:34am, Tanaka Akira wrote:
} Subject: closing stdin (Re: _make suppress error on OpenBSD.)
}
} > coulee% make awk: i/o error occurred while closing /dev/stdin
} 
} This problem is caused by the one true awk behavior.  (OpenBSD comes
} with the one true awk.)  The one true awk closes all file descriptors
} which should be opened when exit and complain if it causes an error.
} 
[...]
} 
} However, we can suppress the error by modifying each awk invocation,
} but I think that completion system should redirect stdin from
} /dev/null instead of closing it.

It's a little more complex than this.

Descriptor 0 is being closed by movefd(0) which is called from zle_main()
before executing any user-defined widget, completion or otherwise.  This
explains why it doesn't work to invoke e.g. vim from a widget function.

This has some odd side effects, including that assorted opens for things
like shared library loading and directory scans for path searching can end
up using FD 0 temporarily (look at strace output).  (Memory mapping of
autoloaded functions also sometimes gets FD 0 but immediately movefd()s it
elsewhere.)

It makes more sense to me from the standpoint of the shell's internals to
simply close FD 0 rather than reopen it on /dev/null.  On the other hand,
I've long forgotten (if I ever knew) why the stdin of ZLE functions is
closed.  (Zefram?)  It's trivial to reopen it again:

	function vi-vi-buffer() {
	    local tmp=${TMPPREFIX}${$}vi
	    print -R - $BUFFER >| $tmp
	    vi $tmp </dev/tty	# Reopen stdin closed by ZLE
	    BUFFER="$(<$tmp)"
	}
	zle -N vi-vi-buffer
	bindkey -v ; bindkey -a v vi-vi-buffer

On the other hand, it's probably a good idea that `exec < file' in a widget
doesn't change the whole shell's standard input, so duping and restoring FD
0 is good even if closing it may not be.

In any case, that latter suggests what I think is the best solution for the
completion system.

Index: Completion/Commands/_complete_help
===================================================================
@@ -3,6 +3,7 @@
 _complete_help() {
   setopt localoptions nullglob rcexpandparam extendedglob
   unsetopt markdirs globsubst shwordsplit nounset ksharrays
+  exec </dev/null	# ZLE closes stdin, which can cause errors
 
   local _sort_tags=_help_sort_tags text i j k
   typeset -A help_funcs help_tags help_sfuncs help_styles
Index: Completion/Core/_main_complete
===================================================================
@@ -18,6 +18,7 @@
 
 setopt localoptions nullglob rcexpandparam extendedglob
 unsetopt markdirs globsubst shwordsplit nounset ksharrays
+exec </dev/null	# ZLE closes stdin, which can cause errors
 
 local func funcs ret=1 tmp _compskip format nm \
       _completers _completer _completer_num curtag _comp_force_list \

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

Re: closing stdin (Re: _make suppress error on OpenBSD.)

[Tanaka Akira]

In article <1000510044838.ZM2434@candle.brasslantern.com>,
  "Bart Schaefer" <schaefer@candle.brasslantern.com> writes:

> In any case, that latter suggests what I think is the best solution for the
> completion system.

Thanks.  I reverted 11280.
-- 
Tanaka Akira