Re: [PATCH v2 1/3] uuid: Add inline helpers to operate on raw buffers

[Christoph Hellwig <hch@lst.de>]

On Wed, Jul 17, 2019 at 05:37:06PM +0200, David Sterba wrote:
> > This entire patch because of BTRFS maintainers, they didn't want the explicit
> > casts. Maybe something has been changed, I dunno.
> 
> No change on our side. The uuids are u8 in the on-disk structures, that
> will stay. The uuid functions use a different type so the casts have to
> be added, that's clear. The question is if it's up to the API to provide
> functions that take u8, or btrfs code to put typecasts everywhere or
> carry own wrappers that do that.

So why do you insist on the u8 for the on-disk format?  uuid_t is
defined in RFC4122 as a stable format, and one of the two origins of
our uuid_t infrastructure is the XFS code, where it is used for the
on-disk format.  What is different in btrfs?

> Specifically for uuid, the endianness might matter, so that we use the
> raw buffers makes things more explicit.

u8 arrays hide the endianess, while the RFC4122 UUID is very clearly
defined as having big endian fields where they are bigger than a byte.