>
>On Sun, 29 Apr 2001 jchacon@genuity.net wrote:
>
>> Can anyone think of a reason to not change CWARNFLAGS to
>> 
>> -Werror -Wall -Wmissing-prototypes -Wstrict-prototypes
>> 
>> instead of
>> 
>>  -Werror -Wreturn-type -Wpointer-arith
>> 
>> which is all that's there today.
>
>Yes. gcc's ability to detect variables used uninitialized is not that
>swift. It generates a lot of false positives.

I didn't say to remove the nounitialized check. I realize that wasn't obvious
but I was only referring to the first line of WARNFLAGS in the Makefile.

>
>> The files so far that broke were bus.h and a few misc ones which either
>> didn't have proper prototyping in scope or in some cases weren't using
>> prototypes. bus.h broke because the definitions for the bus_* functions were
>> all done in K&R style. That's a quick fix.
>> 
>> The main reason here is to catch prototype errors and type promotions/etc
>> that -Wall pulls in. Plus a quick scan of the other port's shows the above
>> options to be more inline with any idea of a standard option set across
>> ports. Hell..the fact that my testing showed possible prototyping/scoping
>> problems tells me this should be there. I've found a few actual bugs so this
>> does seem important to have.
>
>I think a better thing to do is get it so we can lint the kernel. :-) The
>main things that we need are 1) C "routines" for lint for asm routines
>called from C, and 2) fake calls (probably in (1)) to C routines called
>only from asm code.

That too. But, properly prototyping, casting, etc are all important and getting
all the ports to do this correctly is important to consider.

James