qpid-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Alan Conway <acon...@redhat.com>
Subject Re: [c++]: messaging api, uuids and handles
Date Thu, 14 Jan 2010 18:44:27 GMT
On 01/14/2010 11:05 AM, Andrew Stitcher wrote:
> On Thu, 2010-01-14 at 09:06 -0500, Alan Conway wrote:
>> On 01/13/2010 11:04 AM, Andrew Stitcher wrote:
>>> On Wed, 2010-01-13 at 10:52 -0500, Alan Conway wrote:
>>>> On 01/13/2010 09:42 AM, Gordon Sim wrote:
>>>>> I have an outstanding Jira (QPID-2323) to add support for uuids as a
>>>>> value for a Variant in the 'new' API (it is wanted for QMF and seems
>>>>> like it might be a generically useful type).
>>>>> The current Uuid class is in qpid::framing and has exposed dependencies
>>>>> on things I don't want to be part of the API (such as boost::array and
>>>>> qpid::framing::Buffer).
>>>>> My thought therefore is to create a copy of that class in the
>>>>> qpid::messaging namespace, pimpled and cleaned up to avoid the undesired
>>>>> dependencies. Does that sound like the right approach?
>>>> I don't think it should be pimpled, it's a value type and the 16-byte
>>>> representation will never change. I'd suggest a char[16] member and manually
>>>> write the stuff that boost::array provides (copy, assign etc.)
>>> I think it's a toss up between these approaches, but I'd note that
>>> pimpling will allow force you to make the interface explicit (including
>>> the assigns/copies etc.) so that you should easily be able to move to
>>> the other implementation. Although in that case the ABI would change,
>>> but the effective API wouldn't.
>> I think the key point here is that a pimpl has reference semantics rather than
>> value semantics. I think UUID should have value semantics i.e.
>>     Uuid x = ...;
>>     UUid y = x;
>>     // modifing y should not affect x
>> With pimpl, modifying y *will* modify x since y is a reference to the same impl
>> as x,
> Not if you have a copying assignment operator, which I think I included
> above, In that case you get value semantics.
> I agree Uuid should have value semantics, but you _can_ implement with a
> pimpl.

Agreed, since there are no modifying operations other than assignment.
Seems excessive though, for a fixed 16 byte quantity.

Apache Qpid - AMQP Messaging Implementation
Project:      http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org

View raw message