Re: [Barry-devel] WinCE port of Barry

[Toby G. <tob...@re...>]

On 21/07/12 07:46, Chris Frey wrote:
> On Mon, Jul 16, 2012 at 12:10:28PM +0100, Toby Gray wrote:
>> The rest of the commits are a combination of changes to the code to fix
>> warnings or errors from the Microsoft compiler. If the commit messages
>> aren't detailed enough for any of these then let me know and I'll fill
>> in the gaps (and update the commit messages).
> I'm merging in the obvious fixes into master, and the changes that I think
> might need to wait until version 0.19.0 (due to abi changes, etc), I've put
> in a new branch v0.19.0.  Expect that branch to be rebased on master
> periodically.  I'm not quite done yet.

Thanks for looking at this even though I didn't get time to respond to 
your first email. Is the only abi change the tr1 smart_ptr support? If 
so then I was planning on seeing if your suggestion for using boost to 
provide tr1 on WinCE would work, as that would certainly be cleaner.

>
> You should be able to rebase your branch onto the v0.19.0 branch to see
> what I haven't processed yet.
>
>
> Some questions:
>
> 1)
>
> Can you explain why you need a lock() member in scoped_lock?
> The idea behind scoped_lock is to force its use in a scoped section,
> to prevent forgotten unlocks and to make the sections obvious.
>
> Was there a real need for it?

Only for the transferring of interest in a socket, which I'm hoping to 
rework to avoid the need for this.

> 2)
>
> Author: Toby Gray <tob...@re...>
> Date:   Thu Jun 7 15:46:53 2012 +0100
>
>      lib: Fixing incorrect forward declaration of ProbeResult as a struct, it's r
> eally a class.
>
> diff --git a/src/controller.h b/src/controller.h
> index 8e3ba85..41727c7 100644
> --- a/src/controller.h
> +++ b/src/controller.h
> @@ -32,7 +32,7 @@ namespace Barry {
>   
>   // forward declarations
>   class SocketRoutingQueue;
> -class ProbeResult;
> +struct ProbeResult;
>   
>   class PrivateControllerData;
>   
>
> Is this a problem with the Microsoft compiler?  I believe that strictly
> speaking, according to the C++ spec, class and struct can be used
> interchangeably.
It's certainly an issue with the Microsoft compiler. I can't quite 
remember the issue but I think it was causing the mangled names of 
either methods with ProbeResult as a parameter or the type information 
(vtable etc.) for the class to not match up. I'll try to reproduce the 
issue and add a bit more detail to the commit message.

Regards,

Toby