Hello Paul, * Paul Eggert wrote on Thu, Jun 22, 2006 at 01:36:20AM CEST: > Ralf Wildenhues <[EMAIL PROTECTED]> writes: > > > Next, for the transition > > time, it's extremely helpful if the package bootstraps unchanged with > > both 2.59 and 2.60. So far, I think all datarootdir induced changes can > > be written in a backwards compatible way (so they work with 2.59). > > This needs documentation, surely. Perhaps even a separate page in the > manual. The issue is confusing, and this stuff needs to be explained > clearly.
That is what I feared. ;-) > > So how about let's make the silencing also backward compatible, to keep > > the smooth upgrade path? > > Sorry, I don't follow. The proposed AC_DATAROOTDIR_CHECKED macro > cannot be used in configure.ac files intended to be portable to 2.59. Yes, it can. The point is that users don't add AC_DATAROOTDIR_CHECKED but instead they define the macro: AC_DEFUN([AC_DATAROOTDIR_CHECKED]) which will simply be ignored by older Autoconf versions. > It sounds like your scenario is more like this -- am I right? > > * The source package works with 2.59. > > * With 2.60 there are a lot of annoying warnings. > > * Maintainer manually reviews each warning, and fixes the underlying > package so that it works under either 2.59 or 2.60. Right. > (An example > or two? How can the package use datarootdir under 2.59 when 2.59 > doesn't define it?) Well, the usual solution is to add an initialization [EMAIL PROTECTED]@ somewhere. > * Unfortunately, even if the underlying package is fixed, autoconf 2.60 > sometimes issues false alarms Right. > (what's an example of this?). A hand-written Makefile.in that is, say, GNU make specific and uses its `include' mechanism; the definition of datarootdir is done in an included snippet. > So there's > a way to shut off all alarms. (There isn't a way to shut off the alarm > selectively, just for the false alarms that have been manually checked, > because ....?) Right. I guess I'll look into writing... Cheers, Ralf
