Subject: Re: [boost] Heads up: 1.55.0 deadlines approaching
From: Vicente J. Botet Escriba (vicente.botet_at_[hidden])
Date: 2013-09-26 12:05:40
Le 26/09/13 17:46, Daniel James a écrit :
> On 26 September 2013 15:49, Vicente J. Botet Escriba
> <vicente.botet_at_[hidden]> wrote:
>> Could I merge this fix once the regression test have cycled
>>
>> Committed revision 85962. Thread: future::then should join the thread before
>> destruction.
>> https://svn.boost.org/trac/boost/changeset/85962
> Is this a bug fix, or a standards compliance fix? If the former than
> yes, such fixes are fine. If the latter then I'm inclined to say yes,
> but another release manager might be more knowledgeable than me, or
> have another opinion?
It is a bug fix.
>
> Actually, I just had a look at the patch, and I don't think I
> understand it. Calling 'join' in the destructor for
> 'future_async_continuation_shared_state' seems redundant since it
> inherits from 'future_async_shared_state_base' and its destructor
> calls 'join'. Am I missing something?
>
Yes, here is the problem the data associated to the
future_async_continuation_shared_state will be destroyed before the
future_async_shared_state_base would try to join. As the thread function
associated to the continuation uses these variables there is undefined
behavior (a big bug :( ).
Currently the regression test on clang- darwin- asan11 are failing on
some the future::then tests. I hope that this change would fix them.
We can anyway wait until this tester runs tomorrow.
Best,
Vicente