: On 17 Jul 2000, Akim Demaille wrote:
: > | Akim Demaille <[EMAIL PROTECTED]> writes:
: > | > | AC_DEFINE_UNQUOTED($zz, 1, test)
: > | Yes, but this is *_UNQUOTED. If it can't possibly accept a third
: > | argument, why is it documented as doing so? And it *did* work in 2.13.
: >
: > I meant the AC_DEFINE family.
: he's saying AC_DEFINE_UNQUOTED.
Yep, and I meant what I wrote :)
: > It cannot have ever *properly* worked. Pay attention that you're
: > referring to the creation of config.h.in.
: >
: > Maybe AC_DEFINE(`echo FOO`) used to be understood by autoheader, but
: > it was a misfeature. In the general case, this cannot be done, hence
: > it must never work :)
:
: I missed this thread: what is your counter-example?
I know what you are going to say: we don't need autoheader, we should
just spit a cleaned up version of confdefs.h, as your patch does. But
I don't share this opinion, so yes, it is important to run autoheader.
But then, just imagine
AC_DEFINE(FOO$$)
and you'll see what I mean. We need something predictable,
terminating, and M4 computable: literals.
And I find this a very sane attitude: no source depends on an infinite
set of symbols.
