Re: [PATCH] block: Use BIT macro from include/linux/bitops.h

[Jens Axboe <axboe@kernel.dk>]

On 05/20/2015 01:50 PM, Jagan Teki wrote:
> On 21 May 2015 at 01:13, Jens Axboe <axboe@kernel.dk> wrote:
>> On 05/20/2015 01:41 PM, Jagan Teki wrote:
>>>
>>> On 21 May 2015 at 00:52, Jens Axboe <axboe@kernel.dk> wrote:
>>>>
>>>> On 05/18/2015 01:14 PM, Jagan Teki wrote:
>>>>>
>>>>>
>>>>> Replace (1 << nr) to BIT(nr) where nr = 0, 1, 2 .... 31
>>>>
>>>>
>>>>
>>>> I don't like it, I think it hurts readability.
>>>
>>>
>>> What do you mean by don't like, using kernel defined macro instead of
>>> numerical assignments huts readability?
>>
>>
>> In the context of the patch, BIT(0) == (1 << 0) is obvious. But if I just
>> came across BIT(7) in the code, I'd have to check, whereas anyone would
>> immediately know that (1 << 7) is the 7th bit set. Hence, readability is
>> worse, and that's important.
>
> I don't how that BIT(7) is tricky to understand as BIT(0) implies to
> be set 0th bit.
> If understanding of BIT(0) is same like to be as BIT(7) and these were
> simplified
> macro's used most of the code in kernel.

Well of course, if you know what BIT(7) is, you know what BIT(0) is. My 
point is that I don't know what either of them are, I'd have to look it 
up. Whereas anyone would immediately know what (1 << 7) or (1 << 0) is 
without having to look further.


-- 
Jens Axboe