On Fri, Aug 13, 2021 at 01:14:58PM +0300, Andy Shevchenko wrote:
> > Andy, does this incremental diff restore your reviewed-by? The awkward
> > piece of this for me is that it introduces a handful of unnecessary
> > memory copies. See some of the new nsl_get_uuid() additions and the
> > extra copy in nsl_uuid_equal()
> 
> It does, thanks! As for the deeper discussion I think you need to talk to
> Christoph. It was his idea to move uuid_t from UAPI to internal kernel type.
> And I think it made and still makes sense to be that way.
> 
> But if we have already users of uuid_t like you are doing here (without this
> patch) then it will be fine I guess. Not my area to advise or decide.

I'm missing a lot of context here.  But that whole uuid/guid thing is
a little complex:

 - for userspace APIs and on-disk formats a uuid is nothing but a blob
 - userspace historically has its own library to deal with this (libuuid),
   which defines a uuid_t itself.

So instead of trying to build abstractions that somehow word in diferent
software ecosystems I think just treating it as the blob that it is for
exchange makes life easіer for everyone.  It also really makes definitions
of on-disk structures more clear when using the raw bytes instead of a
semi-opaque typedef.

Reply via email to