Brett Lymn said: > Did I say stress anywhere?
You didn't, but you should have. > I said test. Sure, doing a build will > stress the machine but all it really tests is that the tool chain is > functional and that the kernel is functional enough to handle the > build. That's actually pretty functional. > When I said "testing the output binaries" You demonstrated that you didn't know what you were talking about. The compiler is not the issue. The userland binaries are not the issue; they're well tested on big endian, little endian, 32-bit, 64-bit... if they suddenly blow up, it will be because of something that effects everything else. > Yes, I read them. There seems to be a presumption that cross building > will result in badly generated code No, you didn't read them. There's no question about what cross-building *generates*. As Art said, other projects get along with it just fine. The point he raised was that they didn't do as much stress testing. The kernel a) runs longer, continuously, than anything else on the system; and b) contains all the opportunities for architecture-specific bugs. For a userland program that does its job and exits, stress testing is not so big an issue. A memory leak might never matter; a null pointer might not be recycled; there are no CPU registers to worry about resetting correctly, or bus initializitions to do exactly once. -- Matthew Weigel hacker [EMAIL PROTECTED]
