[odb-users] Wrapping existing members with std::atomic<>

Andrew Cunningham odb at a-cunningham.com
Fri Jul 14 10:36:43 EDT 2017


HI Boris,
I already have defined a wrapper for my 'own' smart pointer, so am pretty
clear about the process. I was hoping for something simpler - I think my
solution is to use a transient atomic and copy that to/from a persistent
field "pre persist" and "postload."

Andrew

On Fri, Jul 14, 2017 at 7:25 AM, Andrew Cunningham <odb at a-cunningham.com>
wrote:

> Hi Boris,
> That's correct. I do not need to worry about atomicity during DB
> operations.
> It's used when I am doing multi-threaded operations on the objects after
> they are loaded.
>
> Andrew
>
> On Fri, Jul 14, 2017 at 12:58 AM, Boris Kolpackov <boris at codesynthesis.com
> > wrote:
>
>> Andrew Cunningham <andrew at a-cunningham.com> writes:
>>
>> >                 I need to take some of my 'existing' persistent member
>> > variables (typically int and bool) and wrap them in std::atomic<>
>> without
>> > messing with my schema.
>>
>> I think the "proper" way to handle this would be to define the
>> odb::wrapper_traits specialization for std::atomic, similar to
>> nullable, optional, smart pointers, etc. For an example, see the
>> specialization for boost::optional in libodb-boost (files of
>> interest are optional.options and optional/wrapper-traits.hxx in
>> the odb/boost/ subdirectory). Except in your case it won't be a
>> NULL handler.
>>
>> There is one snag with this approach: the odb::wrapper_traits
>> interface expects you to return references to the underlying
>> value. I think there is a way to make it work if you don't
>> need atomicity during database operations. Is this a reasonable
>> assumption in your case?
>>
>> Boris
>>
>
>


More information about the odb-users mailing list