Re: [configure bug while building] flex beta version 2.5.15

[Bruce Lilly]

Paul Eggert wrote:
> > From: Bruce Lilly <address@hidden>
> > Date: Thu, 29 Aug 2002 11:40:58 -0400
> >
> > Tim Van Holder wrote:
> > > it will only have an effect if interactively running
> >
> > > commands - PS1 never gets evaluated when running a script
> >
> > Not so -- I've seen configure fail unless/until PS1
> > was reset, either externally or by inserting a line
> > near the top of the script -- otherwise it would not
> > have been reported as a configure bug.
>
>
> Do you have more details about that box?  Perhaps a copy of that bug
> report?

That was ksh running under a previous version of
UWIN (http://www.research.att.com/sw/tools/uwin),
probably version 2.9.  Yhe problem does not occur
in the current (3.0) release of UWIN or in a beta
of 3.1, scheduled to be released (literally) within
a matter of days.  The problem may exist in ksh on
other platforms as well.

It was reported as a bug in the bison 1.33 configure
script ca. 22 Feb 2002.

> How "near the top" does the "unset PS1" have to be?

I don't know that unset is effective; certainly
PS1='$ '
as reported in the bison bug report worked in that
configure script.  More recent configure scripts run
subshells, so the export of PS1 should be ensured.

> In other words,
> how does this PS1-related bug affect the execution of shell scripts?
> For example, does the problem occur only if the shell script executes
> something inside backquotes?

As I no longer have access to a system where the problem
is exhibited, it's difficult to answer precisely.  Before
the first use of $? would be a good idea though, for
obvious reasons.