Messages in this thread |  | | Subject | Kernel config testing | Date | Tue, 30 Jul 1996 12:18:54 -0600 | From | Warner Losh <> |
| |
While I was at a previous company, we had to compile a bunch of things with about 20 different options. Rather than compile each and every combination of them, we compiled about 20, one each with the options on. We did this to ensure that most of the time things would compile no matter what options you had. This took care of the case where you had #ifdef CONFIG_FOO ... #else ... #endif and was generally a good thing. What it failed to catch was the if defined this and defined that, but those were relatively rare in the world.
A good project for someone wanting to learn about the kernel would be to build a tool that would take arch/xxx/config.in and then build the kernel with each option, in turn, turned on. This should give a good indication for the common cases where things will break or not. There will likely have to be some hints files, as I've seen, from time to time in the past, cases where you have to have some options enabled for other options to work. These hints could also make sure that some of the more common #if defined(this) && defined(that) configurations were tested as well.
Until someone has the time and energy to undertake this project, automated test builds of the kernel will likely not happen.
Done correctly, I would imagine this would take someone about a month or two to do, assuming they are just learning Linux and its build system. Much less for those that know it in detail.
As to the politics of the situation, I'll leave that for others to flame about in another venue.
Warner
|  |