On Tue, May 18, 2004 at 09:03:44AM +0900, GOTO Masanori wrote: > At Mon, 17 May 2004 14:03:30 -0400, > Daniel Jacobowitz wrote: > > On Sun, May 16, 2004 at 02:17:53AM +0900, GOTO Masanori wrote: > > > I also fully agreed. The latest kernel uses __STRICT_ANSI__ in only > > > types.h and byteorder.h for each architecture. But there are a lot of > > > __s64 and __u64 use without __STRICT_ANSI__ ifdefs in the kernel > > > headers even if that clause is excluded from __KERNEL__. This means > > > that we don't care about this problem. And nowadays we're moving > > > standard to ISO C99, and that includes "long long". > > > > > > The only remained problem is: the default standard of gcc 3.3 is not > > > ISO C99. But "long long" works with even gcc 2.95.3 (which is at > > > least required for kernel 2.6 compilation). So it's not exact > > > problem. > > > > > > Attached patch removes all __STRICT_ANSI__ from the latest kernel > > > 2.6.6 and today's bk. I'll ask it to lkml and put this patch into lkh > > > cvs, if you have no objection. > > > > Have you tried building lkh with this patch? The included testsuite > > will fail unless I'm very confused. If you remove the __STRICT_ANSI__, > > you will need to add strategic uses of __extension__. > > Of course I built it. Could you show me your build log?
I hadn't tried it, but now I have. "-ansi -pedantic" causes "long long" to warn, so this shouldn't work. I just tried it; the patch didn't apply because of a conflict in asm-parisc/byteorder.h. After fixing that the testsuite failed. Note that the testsuite is not run during debian/rules build, only during debian/rules binary; there's a bug filed on CDBS about the reason why. -- Daniel Jacobowitz
