Re: compile problem

[Alexandre Duret-Lutz]

>>> "Gary" == Gary V Vaughan <address@hidden> writes:

[...]

 Gary> I'm not saying that the compile script should be used,
 Gary> just that the tests to determine whether it might be
 Gary> needed are run as a matter of course.  Looking at the
 Gary> implementation of AM_PROG_CC_C_O, I see that it would
 Gary> need a bit of refactoring to make that happen...

Yeah.  Presently you need `compile' if you use the macro, and
people mind about extra useless scripts.

 Gary> My feeling at the moment is that it would be a small price to pay for
 Gary> simplifying the interface to automake (i.e. _not_ being sometimes told,
 Gary> partway through the automake run, that you must change your configure.ac
 Gary> and start again at the aclocal stage).

I agree that being told to update configure.ac can be annoying,
even if it only occur once, but I'm not convinced that we should
call such macros unconditionally.

By this argument it looks like we should also always run
AC_INIT_LIBTOOL so that Automake does not demand it once you
introduce the first variable _LTLIBRARIES variable in a
Makfile.am.  Likewise for AC_PROG_YACC, AC_PROG_RANLIB, etc.

I once thought that since Automake knew what extra macros where
required, it could generate a configure.ac fragment with these.
But that sounds too fragile and tricky.
-- 
Alexandre Duret-Lutz