Re:Typedefs & Linus
by ori_b on Thursday October 14, 2010 @10:33AM
(#1790996)
Sorry, what?
[ Parent
| on HN ]
He's correct that it's legal ANSI C, but pvg is also correct in pointing out that compilers will warn you. (And being legal ANSI C is not necessarily the same thing as being portable.)
Right but it's not about being portable in kernel development -- the entire thing you're building is in and of itself nonportable.
Err, not really. You should be as portable as possible across different architectures. For example, on a 32-bit machine you can get away with using an unsigned int to represent an address, but that will break on a 64-bit machine.
The fact that the Linux kernel works on such disparate architectures also works against your point.
No it's not portable C. We replicate code for different architectures. There are portable sections, true, but the entire kernel isn't portable. Look in the source directory to find the amd64/x86/ARM specific code.
If you know how to make
__asm__ __volatile__(...)
portable I'd love to hear it.
We may be in violent agreement. Large swaths of the kernel can be portable, and you should code them as such. Of course architecture specific parts of the kernel will not be portable.
Yeah I think you're right. :)
It's legal, but taking the address and assigning directly to integers without a cast is anything but idiomatic.