On 09/13/2017 11:29 PM, Greg KH wrote:
> On Wed, Sep 13, 2017 at 09:23:31PM +0200, Lars-Peter Clausen wrote:
>> On 09/13/2017 08:58 PM, Greg KH wrote:
>>> On Wed, Sep 13, 2017 at 06:03:10PM +0100, Jonathan Cameron wrote:
>>>> On Wed, 13 Sep 2017 14:14:07 +0530
>>>> Himanshi Jain <himshijain...@gmail.com> wrote:
>>>>
>>>>> Add __ATTR_NAMED macro similar to __ATTR but taking name as a
>>>>> string instead of implicit conversion of argument to string using
>>>>> the macro _stringify(_name).
>>>>>
>>>>> Signed-off-by: Himanshi Jain <himshijain...@gmail.com>
>>>>> ---
>>>>> include/linux/sysfs.h | 7 +++++++
>>>>> 1 file changed, 7 insertions(+)
>>>>>
>>>>> diff --git a/include/linux/sysfs.h b/include/linux/sysfs.h
>>>>> index aa02c32..20321cf 100644
>>>>> --- a/include/linux/sysfs.h
>>>>> +++ b/include/linux/sysfs.h
>>>>> @@ -104,6 +104,13 @@ struct attribute_group {
>>>>> .store = _store, \
>>>>> }
>>>>>
>>>>> +#define __ATTR_NAMED(_name, _mode, _show, _store) {
>>>>> \
>>>>
>>>> I'm not sure about the naming here. The normal __ATTR macro is also
>>>> 'named'. Maybe something as awful as
>>>>
>>>> __ATTR_STRING_NAME ?
>>>>
>>>> Greg what do you think?
>>>
>>> ick ick ick.
>>>
>>>> This is all to allow us to have names with operators in them without
>>>> checkpatch complaining about them... A worthwhile aim just to stop
>>>> more people wasting time trying to 'fix' those cases by adding spaces.
>>>
>>> Yeah, but this really seems "heavy" for just a crazy sysfs name in a
>>> macro. Adding a whole new "core" define for that is a hard sell...
>>>
>>> I also want to get rid of the "generic" __ATTR type macros, and force
>>> people to use the proper _RW and friends instead. I don't want to add
>>> another new one that people will start to use that I later have to
>>> change...
>>>
>>> So no, I don't like this, how about just changing your macros instead?
>>> No one else has this problem :)
>>
>> Nobody else realized they have this problem yet. E.g. there are a few users
>> of __ATTR in block/genhd.c that have the same issue and are likely to
>> generate the same false positives from static checkers.
>
> Then fix the broken static checkers :)
The static checkers aren't broken, the macro is. It takes a string
parameter, but instead of a string it is passed as an expression and then
transformed to string using preprocessor magic in the macro. That's not very
good semantics. And hence this gets detected as a false positive, because
nobody expects such strange behavior.
