svn commit: r212182 - head/sys/kern

[mdf at FreeBSD.org]

On Mon, Sep 6, 2010 at 6:45 AM, Bruce Evans <brde at optusnet.com.au> wrote:
> On Fri, 3 Sep 2010, pluknet wrote:
>
>> On 3 September 2010 21:23, Matthew D Fleming <mdf at freebsd.org> wrote:
>>>
>>> Log:
>>>  Fix user-space libsbuf build.  Why isn't CTASSERT available to
>>>  user-space?
>
> Well, user headers shouldn't be enlisted to check for kernel bugs that
> can be checked well enough in the kernel.

I agree, but in this case one could define different constants for
user space and kernel space, and the code could be wrong only for
user-space, where there isn't a compile-time assert.

I do always appreciate style and standards advice from Mr Bruce.  It's
the only way to learn (usually, hopefully, from other people's
mistakes ;-)

Thanks,
matthew

>>> Modified:
>>>  head/sys/kern/subr_sbuf.c
>>>
>>> Modified: head/sys/kern/subr_sbuf.c
>>>
>>> ==============================================================================
>>> --- head/sys/kern/subr_sbuf.c   Fri Sep  3 16:12:39 2010        (r212181)
>>> +++ head/sys/kern/subr_sbuf.c   Fri Sep  3 17:23:26 2010        (r212182)
>>> @@ -116,8 +116,10 @@ _assert_sbuf_state(const char *fun, stru
>>>
>>>  #endif /* _KERNEL && INVARIANTS */
>>>
>>> +#ifdef _KERNEL
>>>  CTASSERT(powerof2(SBUF_MAXEXTENDSIZE));
>>>  CTASSERT(powerof2(SBUF_MAXEXTENDINCR));
>>> +#endif
>>>
>>>  static int
>>>  sbuf_extendsize(int size)
>>
>> Hi,
>>
>> as I can see, the next (and maybe preferred) model is used in system
>> headers:
>>
>> #ifdef CTASSERT
>> CTASSERT(...);
>> #endif
>
> Needed, even in the kernel, since CTASSERT() is only defined if the
> kernel-only header <sys/systm.h> has been included.
>
> If this macro were defined in a user header, then it would have to be
> more global (probably defined in <sys/cdefs.h>) but not in the application
> namespace (probably spelled __CTASSERT()), so it would be uglier.
>
> Bruce