On Tuesday 04 September 2007 22:25, Ralf Wildenhues wrote: > FWIW, just to set things a wee bit straighter: Autoconf "chose" to > add the datarootdir indirection exactly because the GNU Coding > Standards required this change. Sure, I realise that; I don't have a problem with it. My point was that, to effect this necessary change, the autoconf developers simply chose to modify certain internal definitions; the obvious solution really, and since the definitions affected had never been formally exposed as a documented interface, a perfectly legitimate change. However, the *unofficial* AC_DEFINE_DIR macro had a built in dependency on these definitions, and became broken as a result of the change; just an illustration of the perils of relying on undocumented behaviour. Of course, my own proposed solution to Vincent's original problem also relied, to a rather lesser extent, on some of the same undocumented features -- specifically that `$prefix' has a default value of `NONE' within `configure', if the user doesn't specify `--prefix=...', and that this default is replaced by the value of `$ac_default_prefix', when any eventual substitution is applied in any AC_OUTPUT file. Sometimes, there is just no way to achieve an objective, without relying on some undocumented feature; when that is necessary, then it's best to keep the dependency to a minimum. Regards, Keith. _______________________________________________ Autoconf mailing list Autoconf@xxxxxxx http://lists.gnu.org/mailman/listinfo/autoconf
