Re: Warning from Boost.Exception

From: Andrey Semashev (andrey.semashev_at_[hidden])
Date: 2019-10-15 09:14:58

On 2019-10-15 12:06, Alexander Grund via Boost wrote:
> 
> Am 15.10.19 um 10:48 schrieb Andrey Semashev via Boost:
>> On 2019-10-15 10:06, Alexander Grund via Boost wrote:
>>>
>>> Where does 'exception.hpp' come from? I'd expect it at 
>>> https://github.com/boostorg/exception/tree/develop/include/boost/exception/exception.hpp 
>>> but it isn't there...
>>
>> It's in https://github.com/boostorg/throw_exception. 
> 
> Then it looks like the warning is indeed correct: 
> https://github.com/boostorg/throw_exception/blob/e2e802e5085e10db7d5ee5ded4c80f663591e976/include/boost/exception/exception.hpp#L168 
> 
> 
> `error_info_container` has a non-virtual dtor although it does have 
> other virtual methods and the derived class has a `delete this` but is 
> not final. Hence the compiler can't be sure that there isn't another 
> derived class.
> 
> Any reason to NOT make the dtor virtual? Performance degrade can be 
> resolved by making the derived class final (where supported which should 
> be almost everywhere nowadays)

The warning is bogus because the object is never destroyed through the 
base class. Marking the destructor virtual would suggest that that is 
possible. And add a virtual call where none is needed.

Marking error_info_container_impl final should help, but we currently 
don't have a macro for that. I created a PR to add one:

https://github.com/boostorg/config/pull/299

After it is merged, Boost.Exception can be updated to take advantage of it.