* Eric Blake wrote on Thu, Oct 23, 2008 at 04:46:39AM CEST: > According to Ralf Wildenhues on 10/22/2008 2:34 PM: > > > > While I haven't looked at this in detail, I do know there are places in > > Autoconf where removing the output files before evaluating the actions > > breaks user scripts. So this sounds quite unsafe to me. > > For that matter, the current refactoring results in more rm processes than > before. Maybe it is worth making _AC_COMPILE_IFELSE_BODY and friends > merely set a shell variable with the list of files to rm, then call a > single rm after the AS_IF has completed the if-pass/if-fail actions of > ac_func_c_try_compile?
Sounds like an idea. > Also, the 'rm -rf conftest.dSYM' is a wasted fork > on most platforms (particularly on those like cygwin where forks are > expensive); should we add a 'test -d conftest.dSYM' first? I guess, although stat isn't all that cheap either. Hmm, could ask Peter whether it would be sufficient to test the first compile (or the $host type?) and then ac_compile_cleanup="rm -rf conftest.dSYM" ... $ac_compile_cleanup You may also want to take note that some scripts use 'break' or even 'break 2' to get out of some tests early; unfortunately, this is even documented in the manual in some places. I see that of course we don't exercise this in our test suite :-/ but its semantics may be relevant for your patches (or not; haven't checked). Cheers, Ralf
