On Tue, 2010-03-02 at 15:48 -0700, Eric Blake wrote: > The change was made back on 2008-11-03 (commit 8e27cc7f5, contained in > autoconf 2.63b), so although it is somewhat of a regression > (AC_LANG_CONFTEST outputs defines now, when it did not do so previously), > there obviously aren't that many packages that try to mix-and-match their > own LANG_CONFTEST with autoconf's LANG_SOURCE, or we might have heard > about this issue sooner.
Indeed, > May I ask why LB_LANG_CONFTEST was created in the first place, Don't really know, TBH. Hysterical raisins I am sure. > rather than > sticking with the autoconf interface, or is that decision lost in history? Yeah. I asked around. Nobody seemed to know, or speak up if they did. I am going to run some tests to see if we can simply use AC_LANG_CONFTEST and be done with the issue. I'd prefer to use as much upstream work as I can (hence my queries about contributing linux kernel tests and macros). I always do. Less surprises like this when you can. > Might it have something to do with the (ancient) bug where we tried to > output #line directives in the original conftest file corresponding to > locations in configure, which was later withdrawn when it was determined > that configure files often exceed 64k lines and it breaks some > preprocessors to have #line numbers that large? Could be, but I really have no idea. b.
signature.asc
Description: This is a digitally signed message part