Jason Thorpe wrote:

> On Fri, 25 Jun 1999 21:54:09 +1000 
>  Simon Burge <simonb@netbsd.org> wrote:
> 
>  > I propose adding a set of macros to <machine/bswap.h> called
>  > htole{16,32,64} and htobe{16,32,64} implemented in terms of bswap* and
>  > no-ops.  There would also be a matching group of {le,be}toh* which do
>  > exactly the same thing.
> 
> This is a wonderful idea :-)  Many device drivers would like this :-)
> 
> Don't forget HTO{BE,LE}{16,32,64}() and {BE,LE}TOH{16,32,64}() too :-)

Those are just the in-place versions, right?

>  > I'm not 100% convinced that <machine/bswap.h> is the right place for
>  > these.  I'd be happier putting these in <machine/endian.h>, but then
>  > that header file would have to include <machine/bswap.h> to get the
>  > bswap* prototypes.  Would doing this break anything?
> 
> <machine/endian.h> is probably the right place, but we need to go on
> a namespace protection sweep in there.

Hmm, I didn't volunteer for that now, did I?  I buggered up my last
namespace test :-)


Thinking about the location of this a little more, it seems that all the
arch-specific <machine/endian.h> files should be little more than:

	#ifndef _MACHINE_ENDIAN_H_
	#define _MACHINE_ENDIAN_H_
	#define _BYTE_ORDER	_{BIG,LITTLE}_ENDIAN

	#include <sys/endian.h>
	#endif /* _MACHINE_ENDIAN_H */

With all the repetitive magic done in <sys/endian.h>.  There's an awful
lot of needless duplication in there at the moment...  It looks like the
i386 and pc532 endian.h would have a little extra goo, but that'd be
about it.  We could even have <sys/endian.h> contain:

	#ifndef _MACHINE_ENDIAN_H
	#include <machine/endian.h>
	#endif

near the top and have in <machine/endian.h>:

	#ifndef _SYS_ENDIAN_H
	#include <sys/endian.h>
	#endif

so that including either <machine/endian.h> or <sys/endian.h> would
DTRT.


Simon.