Lack of recognition of --enable-mule with configure.ac, autoconf 2.62

[Stephen J. Turnbull]

Stephen J. Turnbull writes:
 > Aidan Kehoe writes:

 >  > It’s a (minor) problem with autoconf versions greater than or
 >  > equal to 2.62. See
 >  > http://mid.gmane.org/200812281257.53590.scop@xemacs.org , a
 >  > message I replied to, if it ever makes it to gmane.

Well, it still isn't there, and it's not in the spam-trap.  Nor is
your post.  Seems possible it got spambucketed inadvertantly.  So if
what you wrote is not covered below, you should resend.

 >  > It shouldn't interfere with building XEmacs, but it is something
 >  > that ideally should be resolved, if only to avoid this sort of
 >  > bug report.

At the moment it doesn't interfere, but the authors of autocarp seem
to think of themselves as software style censors, so it could turn
into an error in the future.

What's happening is that there are two sanctioned ways of defining
configure arguments, AC_ARG_WITH and AC_ARG_ENABLE, and any other way
of defining an argument will get the warning, just to make sure you
understand that you are a very bad person indeed!  In particular, the
XE_COMPLEX_OPTION stuff that Malcolm and I implemented apparently
should be considered the work of the very Devil himself.

There are three possibilities:

Quick and dirty:          Invoke AC_DISABLE_OPTION_CHECKING.
Problems:                 No way to be sure the style censor won't remove
                          this feature.
                          AC_REQUIRES will bump from 2.59 to 2.62, I'm
                          afraid.

SSSSSSSSneaky (Gollum!):  Use the undocumented method of declaring
                          options to autoconf in XE_COMPLEX_OPTION.
Problems:                 No way to be sure the style censor won't remove
                          or change this feature.

Good little boy method:   Refactor XE_COMPLEX_OPTION to use
                          AC_ARG_WITH or AC_ARG_ENABLE to get a string
                          argument, then break that down later.
Problems:                 The distinction between AC_ARG_ENABLE and
                          AC_ARG_WITH is bogus.  We currently have a
                          simple hack to make them equivalent, but I
                          suspect this will fail to placate the style
                          SS and the warnings will return.
                          XE_COMPLEX_OPTION is a fragile hack.
                          Refactoring it may break it, with attendant
                          hilarity.

Oh yeah, and there's always

The Right Stuff:          Abandon autotools!
Problems:                 None imaginable, but lots of work. ;-)


So I guess what it comes down to is whether we need to be backward
compatible to pre-FASCIST_OPTION_CHECKING versions of autocarp.
Probably not, since we distribute the configure script.