Re: [PATCH] kobject: Suppress compiler warning with gcc 3.x

[ebiederm@xmission.com (Eric W. Biederman)]

Américo Wang <xiyou.wangcong@gmail.com> writes:

> On Wed, Jun 02, 2010 at 11:02:28PM -0700, Eric W. Biederman wrote:
>>Américo Wang <xiyou.wangcong@gmail.com> writes:
>>
>>> On Thu, Jun 03, 2010 at 07:31:41AM +0200, Eric Dumazet wrote:
>>>>Le jeudi 03 juin 2010 à 14:23 +0900, Tetsuo Handa a écrit :
>>>>> Gcc 3.x generates a warning
>>>>> 
>>>>>   include/linux/sysfs.h:183: warning: parameter has incomplete type
>>>>> 
>>>>> on each file.
>>>>> Suppress the warning by moving the definition of "enum kobj_ns_type"
>>>>> to before "#include <linux/sysfs.h>".
>>>>> 
>>>>> Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
>>>>
>>>>
>>>>I was about to submit same patch, but I was also reverting 27eabc7cb4b3
>>>>(sysfs: Don't use enums in inline function declaration.)
>>>>
>>>>So that sysfs_exit_ns() prototype is consistent regardless of
>>>>CONFIG_SYSFS
>>>>
>>>>What do you think ?
>>>>
>>>
>>> This looks odd, in sysfs.h we do have a forward declaration of
>>> enum kobj_ns_type... I am wondering why gcc 3.x doesn't recognize it.
>>
>>Because the replacement is an inline, and we are passing the enum by value
>>gcc wants to see the full definition, at the point where the inline function
>>is declared.
>>
>
> But sysfs_exit_ns() is not inlined if CONFIG_SYSFS here.

You are right gcc 3.x seems full of it.  I was thinking of the !CONFIG_SYSFS
case where I had to pull out the enum because it is in fact inlined.

If we are going to worry about gcc 3.x and about my !CONFIG_SYSFS
inline function the clean way to handle this seems to be to introduce
a new header file that both sysfs.h and kobject.h can include.
Assuming/requiring that files that include sysfs.h also include
kobject.h to seems like the wrong way to go.

Eric