Re: [21.4,21.5] configure.in (have_dl): Add modules to MAKE_SUBDIR ...

Hrvoje> I wouldn't like to deal with these surprises in a program like Hrvoje> XEmacs. Autoconf caching is a serious pain even in a small program Hrvoje> like Wget where I sometimes had to break its legs just so that some Hrvoje> tests in a loop work. A> Humm... Could you describe the nature of the problems? I just don't A> understand how it turns out that almost no other program except XEmacs A> disables autoconf caching... Autoconf caching is a really really bad idea, except when you're sharing a single set of configuration information among several programs, as is the case, for example, of the "Cygnus toolchain". Even then, it's only for people who know what they're doing. I disabled caching in XEmacs years ago. It was one of the biggest maintainability improvements in the XEmacs configure process. Perhaps my single biggest contribution to free software is convincing the Autoconf people to disable caching in autoconf by default. It took me a couple of tries, and a changing of the guard in the autoconf maintainership. Look for posts by me to the autoconf mailing list for details. This change is in autoconf 2.50, I believe. Short answer: if nothing has changed in the configuration, there's no need to run configure. And if something has changed, you need to re-run every test anyways, because configure can't know what has changed. Alternatively, show me a test that doesn't need to be run again, because you "know" it must give the same result.