Re: subobj8.test fix + am_aux_dir issues

[Alexandre Duret-Lutz]

>>> "Tom" == Tom Tromey <address@hidden> writes:

 >>>>>> "adl" == Alexandre Duret-Lutz <address@hidden> writes:

[...]

 Tom> This patch is ugly, for the reasons listed in the FIXME comment.  Also
 Tom> it doesn't work with an aux dir.  

This is just a matter of using $ac_aux_dir instead of $srcdir, methinks.

[...]

 adl> However, since $ac_aux_dir is safe to use from configure, all we
 adl> have to do is to check script existence or features using
 adl> $ac_aux_dir but use $am_aux_dir for AC_SUBSTed values.

 Tom> I don't know if we can always guarantee this.
 Tom> For instance, in minuso.m4 we couldn't.
 Tom> Someone might plausibly try to use $CC after calling AM_PROG_CC_C_O.

Yeah, I saw your mail to automake@ about this yesterday
and thought "ouch".

 Tom> I guess this is really a constraint on how am_aux_dir can be used.
 Tom> But working with it seems hard.

I agree.

So I'm going back to the original idea: set $am_aux_dir to the
absolute value of $ac_aux_dir (that's what AM_AUX_DIR_EXPAND
does), and use it.  At least for 1.5, or until Autoconf offers a
way to automatically adjust the relativeness of some variables
(as it already does for @INSTALL@ or @top_srcdir@: these
variables are adjusted by config.status for each subdirectory).

I understand it has one drawback: you can't move a configured
tree.  But I'm not sure someone ever did that, because the
way $MISSING is defined prevents this (for the same reason).

[Aside: while playing with $am_aux_dir, I've found that
some Makefiles rely on $MISSING beeing an absolute path.  For
instance Makefile.maint (shared by fetish and autoconf) does
something like this

        cd $(t) && mv $(distdir) $(distdir).old \
          && $(AMTAR) -zxf ../$(distdir).tar.gz

where $(AMTAR) is defined using $MISSING.  This obviously fails
when $MISSING is relative.]

[...]
-- 
Alexandre Duret-Lutz