Hi Ralf, > The remaining sed script could be replaced by AC_SUBST_FILE instances, > although you'd need to modify the gnulib sources then. > > In order to address the fairly large config.status startup overhead, an > alternative would be to just declare the headers AC_CONFIG_FILES, then > they'd be generated normally after configure has run.
I don't believe these two are good advices for gnulib. Both suggestions require to put file names into configure.ac or .m4 files. Each time we did that in the past, it caused *major* trouble that was very hard to get rid of afterwards. Recall the still unresolved story about AC_LIBOBJ [1] and the problem of AC_CONFIG_LINKS [2] that caused us so many spurious bug reports from users who were never running "make distclean"; I believe AC_CONFIG_FILES has the same problem. The good use of autoconf and automake that I've learned the hard way is: +-----------------------------------------------------------------+ | - Use autoconf macros for determining platform dependencies. | | - Use Makefile rules (in Makefile.am files) for creating files. | +-----------------------------------------------------------------+ > > + (cd $(top_builddir) && ./config.status --file=-) < > > $(srcdir)/time.in.h | \ This change is acceptable to me only because it keeps file names out of configure.ac and *.m4. Bruno [1] http://lists.gnu.org/archive/html/bug-automake/2010-05/msg00017.html [2] http://lists.gnu.org/archive/html/bug-gnulib/2010-04/msg00222.html
