Regarding Autotools, and attempts to replace it

In the world of software development on Linux, especially in the open source world, there is a set of packages called "autotools", which abstracts and manages portability between compilers and other aspects of the software build environment. Nobody likes it. In fact, most people who use it hate it. There are many attempts to replace it.

And the problems with autotools are getting worse, because it itself was never designed to have cleanly portable control files between versions. Back when most everyone just FTPed down a tarball, and then ran the prebuilt ./configure, it worked pretty well. Now that people pull a projects raw repo over SVN, BZR, or GIT, and then have to run liptoolize, aclocal, automake, autoconf, etc themselves, and who knows what version of autotools is locally installed, all hell breaks loose.

With respect to all the autotools replacements, such as cmake, Ant, etc, and all the other ones mentioned in Eric Raymond's recent blog post: they ALL are some multiple combination of horribly slow, enforce their own special "one true way" of laying out source trees, are specific to what languages they will deign to handle, have abysmally bad error messages when there is a problem (and being worse than autotools in this respect is an amazing achievement), require the installation of a JVM and a huge pile of buggy poorly documented class files, require the installation of a huge pile of buggy poorly documented Python modules, require the installation of a huge pile of buggy poorly documented Perl modules, cannot intelligently detect and handle optional build dependencies, cannot cross compile, cannot build out of a read-only source tree, cannot build out of tree, and/or cannot build shared object files.

I wish to gods and monsters that this was not true, but it is. And until the writers of the competing build chain systems understand why all this stuff is important, and are willing to support it, autotools will stick around, and people will continue to use it.

This is not to say that it cannot be used better. One of Monty Taylor's herculian tasks the Drizzle project has been pandora build, which is a refactoring and rewrite to the years of cargo-cult accumulated cruft that has infested most autotools based open source projects.

No comments:

Post a Comment