"Greg A. Woods" <woods%planix.com@localhost> wrote: > I would very strongly question the "hundreds of systems" part of that > claim these days, even if one includes every release of every kind of > system where these tests have actually been run "hundreds of times"! The autoconf makros have been tested on more platforms than any other configure-like thing out there. Sure, the resulting configure script might be only tested on a few platfomrms. And sure, developers can (and often do) screw up when writing configure.ac files - but that's not autoconf's fault, because many just write stupid tests without understanding the way autoconf works. > Even worse, many people use autoconf as a crutch and then fail to > write good portable code that doesn't need compile-time configuration > in the first place! This is not an autoconf problem, but a user-problem. If you want to use autoconf, you should first understand the philosophy behind it - testing for features instead of making assumptions based on OS etc. > To that end one should avoid using system identifiers if possible, but > sometimes these are indeed the best way to control compile-time > configuration. System identifiers are NEVER a good way to control compiler-time configuration, _NEVER_! You want to know features, _NOT_ operating systems! The features can and _WILL_ change over time! Just doing #ifdef __NetBSD__ will horribly fail, as features were removed over time and others added. And once a new release comes out, you have to update your code. Good luck updating your code whenever there's a new release of any OS in the world that changes some feature! -- Jonathan
Attachment:
signature.asc
Description: PGP signature