Hi Dan, On Mon, Apr 01, 2002 at 11:38:27AM -0800, Dan Kegel wrote: > Ossama Othman wrote: > > I have to agree with Paul. Sometimes it's just not possible to rely > > on compile/link-time tests alone. > > That's ok if three things hold: > 1. configure.in authors avoid gratuitous use of AC_TRY_RUN.
That should certainly be feasible unless there are some bizarre platforms out there. In any case, this shouldn't be a problem if developers use a reasonable "action-if-cross-compiling" argument for AC_TRY_RUN, such as a compile/link-time test. > 2. All autoconf macros that use AC_TRY_RUN allow overriding the > test via environment variables. (This comes for free if you > use the cache, I think.) configure.in authors who write > custom macros may need education on this. I would tell them to RTFM. :-) > 3. AC_TRY_RUN lets the user specify how to run code on the target, > e.g. with a --with-target-run=foo.sh option, where foo.sh is > a script that runs the given command remotely, e.g. via ssh. > (By default, configure would assume you can't run code on the target > when cross-compiling, just like now.) BTW, did terminology change? I thought that "target" was only used for tool chains (e.g. compiler, linker, etc). Shouldn't the bulk of applications be using "host." For example, "./configure --build=... --host=...". If terminology has changed then someone needs to update the autoconf manual on the GNU autoconf web site. -Ossama -- Ossama Othman <[EMAIL PROTECTED]> Distributed Object Computing Laboratory, Univ. of California at Irvine 1024D/F7A394A8 - 84ED AA0B 1203 99E4 1068 70E6 5EB7 5E71 F7A3 94A8
