Re: [PATCH 1/3] Add a kstrtobool function matching semantics of existing in kernel equivalents.

[Jonathan Cameron <jic23@cam.ac.uk>]

On 03/23/11 16:01, Greg KH wrote:
> On Wed, Mar 23, 2011 at 05:30:11PM +0200, Alexey Dobriyan wrote:
>> On Wed, Mar 23, 2011 at 3:39 PM, Jonathan Cameron <jic23@cam.ac.uk> wrote:
>>> +int kstrtobool(const char *s, bool *res)
>>> +{
>>> +       switch (s[0]) {
>>> +       case 'y':
>>> +       case 'Y':
>>> +       case '1':
>>> +               *res = true;
>>> +       case 'n':
>>> +       case 'N':
>>> +       case '0':
>>> +               *res = false;
>>> +       default:
>>> +               return -EINVAL;
>>> +       }
>>> +       return 0;
>>> +}
>>
>> sigh... such simple thing and so many bugs
Yeah, not by best work.
>>
>> The only values such function should accept is 0 and 1.
> 
> Why?  That's not the way the existing kernel functions that use this
> work.
> 
>> Have you read the rest of kstrto*() code?
>> Where is newline check?
There are plenty of nastier cases that get through than a newline
in the middle of the string (ybobsyouruncle -> 1 nyes->0 :)
>>
>> Anyway, I think it's better do not exist.
> 
> I think it is, as it's already duplicated in at least 2 different places
> in the kernel, and probably more.  Once we get this implementation
> working correctly, we don't need to rewrite it again.
Perhaps naming it like this is a bad idea.  It manages to imply that it
has the same level of strict checking which is seen in the other kstrto*
functions - which is self evidently not true!

The alternative is to try and pin down future interfaces to a narrower set
of 'true' and 'false' values.  We can't realistically change this pair,
(even to 'fix' them) but maybe we can ensure future versions only take 0 or 1?
That sort of simple convention would make life simpler!

Jonathan