Re: [PATCH] sparse: Increase pre_buffer[] and check overflow

[Al Viro <viro@ZenIV.linux.org.uk>]

On Wed, Jul 09, 2008 at 06:32:12PM +0200, Sam Ravnborg wrote:
> So the expalnation seems that gcc for mips define much more
> than the usual gcc does.
> My gcc define 76 symbols for i386.
> 
> And we use this stuff in the kernel.

How much of it do we really use?  Let's see - on i386 gcc-4.1.2 I see
79 symbols.  64 simply never occur in the tree.  At all.  Out of remaining
15, we have

__GNUC__, __GNUC_MAJOR__, __GNUC_PATCHLEVEL__ - provided by sparse.
__STDC__: few users, provided by sparse.
__SIZE_TYPE__: one odd user, defined by sparse anyway
__PTRDIFF_TYPE__: one odd user, defined by sparse anyway

__linux__ - few users, explicitly added in top-level Makefile
linux - 3 users.  Defined in top-level Makefile.
unix - no real users (some instances, of course, but none outside of comments,
#include pathnames and string constants).  Defined in top-level Makefile,
anyway.

__USER_LABEL_PREFIX__, __REGISTER_PREFIX__ - arch/m68knommu/lib/*.S; not
a sparse fodder anyway *and* defaults are given in files themselves.

__ELF__: arch/alpha/boot/tools/objstrip.c (userland helper, actually, *and*
misplaced there; it's used as a proxy for type of kernel image)

__i386__: a bunch
__i386: one user, redundant (__i386__ *and* i386 in the same #if)
i386: 3 users besides the aforementioned one.

So...  Only 3 symbols out of the entire bunch are arch-dependent *and* not
provided by sparse itself.  Absolute majority of the rest is never ever
used in the tree.

I very much doubt that mips situation is seriously different...