Zack's Kernel News

An In-Kernel C Compiler?

Ingo Molnar suggested that the kernel should include its own source compiler inside the codebase itself, a tightly integrated Linux alternative to GCC that would eliminate the nightmare conflicts that have occasionally raged between the kernel developers and the GCC developers. If anyone less involved in kernel development than Ingo suggested something like this, I suspect it would be met with a high degree of skepticism. In fact, even with Ingo suggesting it, there was some of that.

His idea was to combine a precompiler, compiler, assembler, and linker into a single tool that would ship with the kernel sources and remain "in lockstep" with each other, avoiding not just the developer conflicts, but also a lot of jumping through hoops that the kernel currently must do to accommodate GCC. As a first step, the tool could simply perform pre-processing that would be fed to GCC. Then gradually more compiler functionality could be added. This, he said, would have an immediate benefit in terms of simplifying the kernel code.

Initially, there was some support for the idea. Steven Rostedt raised the issue, suggesting that Ingo's idea was a good one; then Ingo replied with this elaboration. And Anton Ertl pointed out that an alternative to GCC wouldn't just help the kernel get around GCC problems, it would help lots of user-space software as well. But he thought forking GCC as a starting point had some good arguments in its favor, including the ability to compile code for a wide variety of hardware architectures. David S. Miller pointed out that one drawback to writing a new compiler would be losing GCC's efficiency at doing preprocessing and compilation all within the same binary, instead of passing data through a pipe. He also didn't like Ingo's idea in general. Instead of working on a compiler, he thought people should focus on writing kernel code. He also thought the whole project would take a lot more time and effort than Ingo expected. Having written a compiler before, Eric W. Biederman also confirmed that it would take a really long time, but he thought Ingo's idea might be worthwhile if it would result in big advances in debugging and compiling speed.

[...]