Re: bison-1.29c 'configure' problems on Solaris 8.

[Paul Eggert]

> From: Akim Demaille <address@hidden>
> Date: 04 Oct 2001 17:31:18 +0200

> First let's find a portable LINENO, *then* move to another shell.

But the attempt to find a portable LINENO is not cost-free.  It is
broken now, and it will take time to fix it, and this is time that
will be almost entirely wasted in practice once we execute configure
with a portable shell.  These days, almost nobody runs on systems that
lack portable shells.  I wouldn't spend a lot of time worrying about
these rare, ancient systems, particularly as the LINENO feature is not
essential for proper operation of "configure".

> But, I would like to see that handled in M4sh, *not* Autoconf.

I don't fully understand this distinction.  How would you implement
the change differently for M4sh?

My change is admittedly untested and probably has several bugs.  I'm
not wedded to it.

But my point is that autoconf 2.52e does not work in Solaris 8 and
probably many other hosts, due to this attempt to support LINENO on a
few ancient hosts.  We shouldn't let our desire to add an unnecessary
nicety on rare, ancient hosts add significant risk of breaking
Autoconf on a lot of much more popular modern hosts.


> Also, I would like to provide means to specify what are the
> requirements you'd like to see fulfilled by the new shell (e.g.,
> functions).

We could add those features as necessary, but I don't think they'll be
needed, at least not at first.  What we want, and what most Autoconf
users want, is a standard environment, and POSIX is clearly the
standard.  I don't think too many scripts rely on extensions to POSIX;
those that do typically put "#!/bin/bash" at the top or something like
that.  This issue can be addressed later, if needed.

> Agreed, I have been tempted by that thinking several times, but I
> stepped back thinking that machine where lineno is not supported are
> probably machine where we will have problems, hence, where line nums
> are the most useful.

If there were a large number of such machines, I would agree with you.
But nowadays such machines are so rare that they're not worth spending
a lot of time to support.

Right now, I am bogged down with Autoconf -- I can't test Autoconf on
my desktop Solaris box, because of the multiple problems introduced
with this LINENO change.  I just don't see why it's worth the
portability hassle of adding this feature.

> What are you referring to?  If configure.lineno exits somewhere, how
> could configure survive?

Configure invokes configure.lineno with something like this:

. ./configure.line
exit 0

But suppose configure.lineno falls off the end, without exiting.  Is
this really possible?  I don't know.  But what is that " 0" doing
there if not?