Re: 1.5.22 fails to configure on AIX 5.1

[Ralf Wildenhues]

[ Cc:ing autoconf-patches; this is
http://thread.gmane.org/gmane.comp.gnu.libtool.general/7105 ]

Hi Howard,

Again several bugs for the price of one bug report.

* Howard Chu wrote on Fri, Jan 13, 2006 at 10:30:07AM CET:
> 
> login shell and $SHELL are /bin/sh. The significant failure follows. 
> Note that al_cr_alnum doesn't get set, that whole chunk of script is 
> corrupted. Likewise later for al_cr_letters. As a minor note, notice 
> that the error message from the unrecognized "set -o posix" doesn't get 
> directed to /dev/null.

Hmm.  The AIX 5.1 /bin/sh has a frightening redirection bug when it
comes to `set':

$ if (set -o posix) >/dev/null 2>&1; then set -o posix; fi

will hang the shell in interactive mode.  Ouch.  Removing the `2>&1'
removes the hang.  Likely, noninteractive mode could mess up some shell
internals as well, I don't know.  Over here, the script finishes though.

Likewise, I have been unable to reproduce any issues _without_ `sh -vx'.
Does it succeed for you then?  If not, what are the failures then?


Darn, this is another issue introduced by my accidental use of CVS
Autoconf/Automake for 1.5.22.  List is growing .. we shouldn't put off
another release too long (one that uses released autotools again). :-/

OTOH, this gets us a bug report against CVS Autoconf:

Autoconf's M4SH_INIT (_AS_BOURNE_COMPATIBLE) used to output
[...]
| elif test -n "${BASH_VERSION+set}${KSH_VERSION+set}" && (set -o posix) 
>/dev/null 2>&1; then
|   set -o posix
| fi

but we changed that to
| elif (set -o posix) >/dev/null 2>&1; then
|   set -o posix
| fi

because, for example, OpenBSD uses a modified pdksh as /bin/sh, and one
of the modifications consists of not setting KSH_VERSION any more.  :-/
I am at a loss as how to detect this cleanly, suggestions welcome.

Cheers,
Ralf