Re: CVS commit: src/sys/modules/examples/fopsmapper

[Kamil Rytarowski <n54%gmx.com@localhost>]

On 01.04.2020 16:10, Paul Goyette wrote:
> On Wed, 1 Apr 2020, Kamil Rytarowski wrote:
>
>> On 01.04.2020 15:47, Robert Elz wrote:
>>>     Date:        Wed, 1 Apr 2020 11:45:53 +0000
>>>     From:        "Kamil Rytarowski" <kamil%netbsd.org@localhost>
>>>     Message-ID:  <20200401114554.05167FB27%cvs.NetBSD.org@localhost>
>>>
>>>   | Log Message:
>>>   | Avoid comparison between signed and unsigned integer
>>>   |
>>>   | Cast PAGE_SIZE to size_t.
>>>
>>> This kind of pedantry is going way too far, PAGE_SIZE is a compile
>>> time constant (1 << PAGE_SHIFT) which is an int (and so signed,
>>> nominally) but one which is known to be positive.
>>>
>>
>> I got reports that certain ports no longer build due to:
>>
>> src/sys/modules/examples/fopsmapper/fopsmapper.c:118:11: error:
>> comparison between signed and unsigned integer expressions
>> [-Werror=sign-compare]
>>  if (size != PAGE_SIZE)
>>           ^~
>> cc1: all warnings being treated as errors
>
>
> There's a lot of modules that fail for this with WARNS=5 when being
> built as loadable modules.
>
> That's why so many of the individual module Makefiles have explicit
> WARNS=4.  (It seems that when modules are built-in as part of the
> kernel, they are by default built with WARNS=4.)
>

I tracked it to WARNS=3

# Rather than our usual WARNS=5, we need to use 3, since there are a
# lot of signed-vs-unsigned compares

WARNS=  3

Personally I have got no preference. I'm fine with an explicit cast as
it is in a single place only needed (hopefully).