cvs commit: src/sys/vm vm_map.c

[Bruce Evans]

On Mon, 28 Jun 2004, Andrew Gallatin wrote:

> Brian Fundakowski Feldman writes:
>  >
>  > GCC not actually type-checking here is certainly a bug from my
>  > viewpoint and one whose existence makes me more than a little
>  > worried about further problems with min() and friends (ones that
>  > I did not introduced ;).  Are there any GCC experts that can help
>  > out, here?
>  >
>
> -Wconversion would find this..
>
>        -Wconversion
>               Warn  if  a prototype causes a type conversion that is different
>               from what would happen to the same argument in the absence of  a
>               prototype.  This includes conversions of fixed point to floating
>               and vice versa, and conversions changing the width or signedness
>               of  a  fixed  point argument except when the same as the default
>               promotion.
>
> However its a little bit too restrictive, as it complains about any
                ^^^^^^ big :-)
> mismatch and the compile doesn't get even one C file done before
> erroring out.
>
> If I add -Wconversion to my cflags, it first trips on sys/bus.h
> (harmless passing of "1" to a function taking a long),
> and then it pukes all over atomic.h:
>
> In file included from ../../../sys/systm.h:41,
>                  from ../../../cam/cam.c:35:
> machine/atomic.h: In function `atomic_set_acq_8':
> machine/atomic.h:308: warning: passing arg 2 of `atomic_set_8' with different width due to prototype
> machine/atomic.h: In function `atomic_set_rel_8':
> machine/atomic.h:308: warning: passing arg 2 of `atomic_set_8' with different width due to prototype
>
> Any idea what its problem is?

This is a FAQ, but I couldn't find the answer in gcc.info.  The problem
is that -Wconversion doesn't do what you want.  It was intended for
discovering portability problems in K&R code and/or old code that doesn't
always have prototypes in scope.  The default promotions can work, but
not if there is a prototype that sometimes forces wider args.  I forget
if the signedness warning was there originally.  Signedness mismatches for
parameters are subtler than size mismatches and tend to get "fixed" by 2
bogus conversions canceling each other.

Bruce