* Philip A. Prindeville wrote on Tue, Jul 08, 2008 at 11:44:08PM CEST: > Ralf Wildenhues wrote: >> If you have a way to actually execute code on the compile >> system (get binfmt to execute 'elksemu $binary' should be >> possible), then you can avoid compile-only tests and make >> AC_COMPUTE_INT use the cheaper execution test. > What about using 'nm'? Generate an object file (or intermediate object > file) with two adjacent symbols (packed) of the same size, and subtract > the difference in addresses? Well, having to rely on 'nm' to even exist is a bit of a pain. For example, the bin86 package that is used along with bcc has the program named nm86, which a tool check would not have found automatically. Then, the nm output format is poorly standardized (see Libtool), relying on the order a compiler outputs symbols is AFAIK not safe. Plus, packing would lead to the wrong solution (AC_CHECK_SIZEOF is intended to return the result that sizeof would, not that sizeof of the packed version of the type would). The proposed ideas of fixing AC_LANG_BOOL_COMPILE_TRY(C) may be pursued. Or bcc fixed. Or you may be able to get by with binfmt_misc (this is actually described in a README for elks-libc). There are however more issueswhen using that. For one, the Autoconf testsuite test 172: C keywords causes bcc to get stuck in an endless loop outputting an error, which just managed to fill my work partition... Cheers, Ralf _______________________________________________ Autoconf mailing list Autoconf@xxxxxxx http://lists.gnu.org/mailman/listinfo/autoconf
