[Bug: 21.5-b27] startup.el "borks" filepath?

[Michael Sperber]

"Stephen J. Turnbull" <stephen at xemacs.org> writes:

> OK, I've replicated this.  I set up an empty install directory in
> ~/usr/local, configured --with-prefix=yes --prefix=~/usr/local, set
> EMACSDEBUGPATHS=YesPlease, then ran make; make install.  I can see
> startup.el change the lisp-directory from the run-in-place
> lisp-directory to the installed lisp-directory, which doesn't exist
> yet.  The first time through, this doesn't happen until make install,
> but once the directories have been created, an ordinary make will
> cause the problem at loadup.el.

Darn.  I can't reproduce it.  On which step in the sequence exactly
does it fail?

> It looks like the installation code creates this directory, then when
> we get to the point of reloading XEmacs pre-installation,
> `paths-for-each-emacs-directory' will find configure-lisp-directory,
> and use that as the preferred value even though it's not yet
> installed.  Worse, for some reason it doesn't find any run-in-place
> lisp.

You're right in that you shouldn't have `configure-lisp-directory' set
to anything.  Do you?  (Robert, do you?)

> I suspect that this is in part happening because autoconf has decided
> to force datadir to be in ${prefix}/share and the logic we use to
> determine whether configure-lisp-directory was user-specified doesn't
> work right, because it expects datadir to be ${prefix}/lib.

No, we have this totally brain-damaged code:

if test "x$datadir" != 'x${prefix}/share'
then
  AC_DEFINE(INFODIR_USER_DEFINED)
  AC_DEFINE(LISPDIR_USER_DEFINED)
  AC_DEFINE(MODULEDIR_USER_DEFINED)
  AC_DEFINE(ETCDIR_USER_DEFINED)
  AC_DEFINE(DOCDIR_USER_DEFINED)
  AC_DEFINE(ARCHLIBDIR_USER_DEFINED)
else
  datadir='${prefix}/lib'
fi

That would set LISPDIR_USER_DEFINED to something if datadir were out
of whack.  But as it isn't specified explicitly, it shouldn't.  (And
it isn't for me.)

-- 
Cheers =8-} Mike
Friede, V?lkerverst?ndigung und ?berhaupt blabla