Hi Ricardo,
On Wed, Sep 20, 2023 at 02:19:54PM +0200, Ricardo Ribalda wrote:
> Hi Laurent, Hi Sakari
>
>
>
> On Tue, 19 Sept 2023 at 23:02, Laurent Pinchart
> <laurent.pinchart@xxxxxxxxxxxxxxxx> wrote:
> >
> > On Tue, Sep 19, 2023 at 08:22:57PM +0000, Sakari Ailus wrote:
> > > Hi Ricardo,
> > >
> > > On Tue, Sep 19, 2023 at 10:10:41PM +0200, Ricardo Ribalda wrote:
> > > > Hi Sakari
> > > >
> > > > On Tue, 19 Sept 2023 at 22:07, Sakari Ailus <sakari.ailus@xxxxxx> wrote:
> > > > >
> > > > > Hi Ricardo,
> > > > >
> > > > > Thanks for the patch.
> > > > >
> > > > > On Tue, Sep 19, 2023 at 04:01:23PM +0200, Ricardo Ribalda wrote:
> > > > > > mipsel64el, ppc64el, ia64, ppc64, sparc64 and x32 have different lenghts
> > > > > > for long long ints, which result in some compilation errors.
> > > > > >
> > > > > > Lets add some castings to help the compiler deal with this.
> > > > > >
> > > > > > We cannot use the Format macro constants ffrom inttypes because they
> > > > > > seem to not be compatible with kernel (__u64 et al) types.
> > > > > >
> > > > > > Signed-off-by: Ricardo Ribalda <ricardo@xxxxxxxxxxx>
> > > > > > ---
> > > > > > yavta.c | 35 +++++++++++++++++++++--------------
> > > > > > 1 file changed, 21 insertions(+), 14 deletions(-)
> > > > > >
> > > > > > diff --git a/yavta.c b/yavta.c
> > > > > > index d562863..bf54e4f 100644
> > > > > > --- a/yavta.c
> > > > > > +++ b/yavta.c
> > > > > > @@ -1313,7 +1313,8 @@ static void video_query_menu(struct device *dev,
> > > > > > printf(" %u: %.32s%s\n", menu.index, menu.name,
> > > > > > menu.index == value ? " (*)" : "");
> > > > > > else
> > > > > > - printf(" %u: %lld%s\n", menu.index, menu.value,
> > > > > > + printf(" %u: %lld%s\n", menu.index,
> > > > >
> > > > > Could you instead use PRId64 for this? You can avoid casting to another
> > > > > type this way. Same for the other cases.
> > > >
> > > > Already tried this:
> > > >
> > > > @@ -1313,7 +1313,7 @@ static void video_query_menu(struct device *dev,
> > > > printf(" %u: %.32s%s\n", menu.index, menu.name,
> > > > menu.index == value ? " (*)" : "");
> > > > else
> > > > - printf(" %u: %lld%s\n", menu.index, menu.value,
> > > > + printf(" %u: %" PRId64 "%s\n", menu.index, menu.value,
> > > > menu.index == value ? " (*)" : "");
> > > > };
> > > > }
> > > >
> > > > but gcc was not happy:
> > > >
> > > > yavta.c: In function ‘video_query_menu’:
> > > > yavta.c:1316:11: warning: format ‘%ld’ expects argument of type ‘long
> > > > int’, but argument 3 has type ‘__s64’ {aka ‘long long int’}
> > > > [-Wformat=]
> > > > 1316 | printf(" %u: %" PRId64 "%s\n", menu.index, menu.value,
> > > > | ^~~~~~~~~ ~~~~~~~~~~
> > > > | |
> > > > | __s64 {aka
> > > > long long int}
> > > > In file included from yavta.c:26:
> > > > /usr/include/inttypes.h:57:34: note: format string is defined here
> > > > 57 | # define PRId64 __PRI64_PREFIX "d"
> > >
> > > I guess I should have expected this...
> > >
> > > I'm not sure if it'd be prettier but another option is to use the PRI*
> > > macros and explicitly cast to a standard type.
>
> I would like to avoid
>
> printf(" Hello %" PRId64 "\n", (uint64_t) value_s64);
>
> That looks very bad :)
I actually prefer this. It doesn't look bad either IMO, apart from the PRI*
macros that are always ugly, but most importantly you're explicitly using
64-bit types that work everywhere.
>
> I believe the current casting is the least of the two evils.
>
>
> > >
> > > Using the standard types in the V4L2 header would have avoided this issue.
> > > I wonder if there's anything to be gained by using the kernel types.
> >
> > The kernel has defined __s64 as signed int int for a long time now, on
> > all architectures, at least since
> >
> > commit 0c79a8e29b5fcbcbfd611daf9d500cfad8370fcf
> > Author: Geert Uytterhoeven <geert@xxxxxxxxxxxxxx>
> > Date: Thu Jan 23 15:53:43 2014 -0800
> >
> > asm/types.h: Remove include/asm-generic/int-l64.h
> >
> > which was merged in v3.14.
> >
> > According to https://buildd.debian.org/status/package.php?p=yavta,
> > however, __s64 seems to be defined as long int on some platforms.
> >
> > /me is puzzled
>
> It does not help that __s64 is long long int and PRId64 is "d". I
> guess we can say that inttypes and kernel types do not play along?
I haven't encountered this issue in the past but I also haven't tried
compiling for odd architectures.
>
> I guess we need kerntypes.h with proper KPRId64 but that is probably
> out of scope here.
This could even depend on the compiler.
I wonder why we aren't using
typedef uint64_t __u64;
in kernel UAPI headers instead. Including inttypes.h should not be an issue
in 2023 anymore.
This problem is certainly wider in scope than yavta.
--
Regards,
Sakari Ailus
