LLVM in the FreeBSD Toolchain

The most obvious incentive for the FreeBSD project to switch from GCC to Clang was the decision by the Free Software Foundation to switch the license of GCC to version 3 of the GPL. This license is unacceptable to a number of large FreeBSD consumers. Given this constraint, the project had a choice of either maintaining a fork of GCC 4.2.1 (the last GPLv2 release), staying with GCC 4.2.1 forever, or switching to another compiler. The first option might have been feasible if other GCC users had desired the same and the cost could have been shared. The second was an adequate stopgap, but the release of the C11 and C++11 specifications— both unsupported by GCC 4.2.1—made this an impossible approach for the longer term. The remaining alternative, to find a different compiler to replace GCC, was the only viable option. The OpenBSD project had previously investigated PCC, which performed an adequate job with C code (although generating less optimised code than even our old GCC), but had no support for C++. The TENDRA compiler had also been considered, but development had largely stopped by 2007. The remaining alternative was Clang, which was still a very young compiler in 2008, but had some significant commercial backing from companies including Apple and Google. In 2009, Roman Divacky and Pawel Worach begin trying to build FreeBSD with Clang and quickly got a working kernel, as long as optimisations were disabled. By May 2011, Clang was able to build the entire base system on both 32-bit and 64-bit x86 and so became a viable migration target. A large number of LLVM Clang bugs were found and fixed as a result of FreeBSD testing the compilation of a large body of code.