Re: [Boost-users] [Review] UUID library (mini-)review starts today, November 23rd

Subject: Re: [Boost-users] [Review] UUID library (mini-)review starts today, November 23rd
From: Andy Tompkins (atompkins_at_[hidden])
Date: 2008-12-06 14:44:11

On Tue, 2 Dec 2008 17:26:13 -0600, "Christian Holmquist"
<c.holmquist_at_[hidden]> said:
> 2008/12/2 Andy Tompkins <atompkins_at_[hidden]>
>
> > > size_type size() The underlying container is boost::array,
> > > shouldn't this function be made static? Or is the size of the uuid
> > > implementation defined and the user shouldn't count on it's
> > > length?
> >
> > It could easily be made static.  It will _always_ return 16.
>
> I think that should be part of the documented interface. It's just
> a matter of not confusing the reader that the uuid may be of
> variable length.

I will document this.

< snip >

> > > Maybe this is out of the question to most, but is the uuid class
> > > needed at all? I'd be happy to see the following working (cause
> > > then I can use our own uuid class with the algorithms provided by
> > > the library).
> > >
> > > #include <boost/uuid/random_generator.hpp> #include
> > > <boost/uuid/support/array.hpp> #include <boost/uuid/io.hpp>
> > >
> > > typedef boost::array<char, 16> id_type;
> > > boost::uuid::random_generator<id_type, boost::mt19937>
> > > uuid_gen(...); my_id id = uuid_gen(); std::cout <<
> > > boost::uuid::format(id) <<std::endl;
> >
> > I not keen on this.  The fact that boost::uuid is implemented with
> > boost::array is just an implementation detail.  That may change.  I
> > want the public interface to remain the same.
>
> Ok. For the sake of interface, does it make sense to you adding const
> char* uuid::data() const and static std::size_t uuid::size() that you
> won't change in future releases?

Hmm, I'm not sure.  There is another post that I haven't got to
replying to that talks about platforms with different sized types
(char, int, ...). If I make sure that sizeof(boost::uuids::uuid) == 16
bytes on every platform then some platforms may get a performance
penalty.  On a platform where sizeof(char) != 1 or sizeof(int) != 4
then boost::uuids::uuid will have to do more work to put bits where
they belong.

Then again, I don't think I can actually guarantee on every platform
that sizeof(boost::uuids::uuid) == 16.

What I have done is make sure that each of the 16 blocks of bits in a
boost::uuids::uuid is at least 8 bits.  Thus boost::uuids::uuid::begin()
and boost::uuids::uuid::end() will walk the 16 blocks correctly.

I know this didn't really answer your question.  I would like to offer
const char* uuid::data() const, but I'm not sure that I can.

> I think it's an important feature for a uuid to store its 128bit value
> as a continues range of bytes, and that users can rely on using it as
> such where needed. Also, that the ordering of the bytes won't change
> in future releases. Uuids are likely to be persistent somewhere, and
> not necessarily trough the means of boost.serialization where version
> data can be attached.

I agree.  I did provide begin() and end() with this purpose in mind.
And the order will never change.

> Regards, Christian

Regards,
Andy Tompkins