[KJ] how about cleaning up all the "__attribute__" defines?

[KJ] how about cleaning up all the "__attribute__" defines?

[Robert P. J. Day]

  it strikes me that a useful cleanup would be to centralize all of
the shortcut "__attribute__" definitions in one place.  consider:

$ grep -r "#define.*attribute.*packed" .
./drivers/scsi/u14-34f.c:#define PACKED__attribute__((packed))
./drivers/scsi/BusLogic.h:#define PACKED __attribute__((packed))
./drivers/scsi/gdth_ioctl.h:#define PACKED  __attribute__((packed))
./drivers/scsi/ultrastor.c:#define PACKED __attribute__((packed))
./drivers/media/dvb/frontends/bcm3510_priv.h:#define PACKED __attribute__((packed))
./include/linux/eventpoll.h:#define EPOLL_PACKED __attribute__((packed))
./include/linux/cycx_x25.h:#define PACKED __attribute__((packed))
./fs/hfsplus/hfsplus_raw.h:#define __packed __attribute__ ((packed))
./fs/befs/befs_fs_types.h:#define PACKED __attribute__ ((__packed__))
./fs/hfs/hfs.h:#define __packed __attribute__ ((packed))

  i mean, seriously, how many different files need to redefine the
macro "PACKED"?  (or, as it were, "__packed"?)  i think it would be
handy to have all of those shortcut defines in one file, with a little
documentation to go along with each so that newbies would know:

  1) what they mean, and
  2) under what circumstances they should be used

thoughts?

rday

_______________________________________________
Kernel-janitors mailing list
Kernel-janitors@lists.osdl.org
https://lists.osdl.org/mailman/listinfo/kernel-janitors

Re: [KJ] how about cleaning up all the "__attribute__" defines?

[walter harms]

Robert P. J. Day wrote:
>   it strikes me that a useful cleanup would be to centralize all of
> the shortcut "__attribute__" definitions in one place.  consider:
> 
> $ grep -r "#define.*attribute.*packed" .
> ./drivers/scsi/u14-34f.c:#define PACKED__attribute__((packed))
> ./drivers/scsi/BusLogic.h:#define PACKED __attribute__((packed))
> ./drivers/scsi/gdth_ioctl.h:#define PACKED  __attribute__((packed))
> ./drivers/scsi/ultrastor.c:#define PACKED __attribute__((packed))
> ./drivers/media/dvb/frontends/bcm3510_priv.h:#define PACKED __attribute__((packed))
> ./include/linux/eventpoll.h:#define EPOLL_PACKED __attribute__((packed))
> ./include/linux/cycx_x25.h:#define PACKED __attribute__((packed))
> ./fs/hfsplus/hfsplus_raw.h:#define __packed __attribute__ ((packed))
> ./fs/befs/befs_fs_types.h:#define PACKED __attribute__ ((__packed__))
> ./fs/hfs/hfs.h:#define __packed __attribute__ ((packed))
> 
>   i mean, seriously, how many different files need to redefine the
> macro "PACKED"?  (or, as it were, "__packed"?)  i think it would be
> handy to have all of those shortcut defines in one file, with a little
> documentation to go along with each so that newbies would know:
> 

It would be useful indeed.
IMHO they should not be used at all except the author knows what he does.
simply let the compiler do its work.
Different architectures have different needs (a real driver problem), so it would
provide a central place to disable these suckers.

re,
 wh



>   1) what they mean, and
>   2) under what circumstances they should be used
> 
> thoughts?
> 
> rday
> 
> _______________________________________________
> Kernel-janitors mailing list
> Kernel-janitors@lists.osdl.org
> https://lists.osdl.org/mailman/listinfo/kernel-janitors
> 
> 
> 
_______________________________________________
Kernel-janitors mailing list
Kernel-janitors@lists.osdl.org
https://lists.osdl.org/mailman/listinfo/kernel-janitors

Re: [KJ] how about cleaning up all the "__attribute__" defines?

[Robert P. J. Day]

On Wed, 17 Jan 2007, walter harms wrote:

>
>
> Robert P. J. Day wrote:
> >   it strikes me that a useful cleanup would be to centralize all of
> > the shortcut "__attribute__" definitions in one place.  consider:
> >
> > $ grep -r "#define.*attribute.*packed" .
> > ./drivers/scsi/u14-34f.c:#define PACKED__attribute__((packed))
> > ./drivers/scsi/BusLogic.h:#define PACKED __attribute__((packed))
> > ./drivers/scsi/gdth_ioctl.h:#define PACKED  __attribute__((packed))
> > ./drivers/scsi/ultrastor.c:#define PACKED __attribute__((packed))
> > ./drivers/media/dvb/frontends/bcm3510_priv.h:#define PACKED __attribute__((packed))
> > ./include/linux/eventpoll.h:#define EPOLL_PACKED __attribute__((packed))
> > ./include/linux/cycx_x25.h:#define PACKED __attribute__((packed))
> > ./fs/hfsplus/hfsplus_raw.h:#define __packed __attribute__ ((packed))
> > ./fs/befs/befs_fs_types.h:#define PACKED __attribute__ ((__packed__))
> > ./fs/hfs/hfs.h:#define __packed __attribute__ ((packed))
> >
> >   i mean, seriously, how many different files need to redefine the
> > macro "PACKED"?  (or, as it were, "__packed"?)  i think it would be
> > handy to have all of those shortcut defines in one file, with a little
> > documentation to go along with each so that newbies would know:
> >
>
> It would be useful indeed.
> IMHO they should not be used at all except the author knows what he does.
> simply let the compiler do its work.
> Different architectures have different needs (a real driver problem), so it would
> provide a central place to disable these suckers.

i did eventually notice that a lot of those shortcuts are defined in
the header files include/linux/compiler-*.h.  but i'm wondering if
those files are entirely up-to-date.  and i think it would still be
easier to follow what's defined if the attributes were defined in a
single header file, rather than spread across more than one.  but
that's just me.

as a starting point, it would be useful to just agree on what
shortcuts should be defined.

rday
_______________________________________________
Kernel-janitors mailing list
Kernel-janitors@lists.osdl.org
https://lists.osdl.org/mailman/listinfo/kernel-janitors