This is the mail archive of the
gcc-help@gcc.gnu.org
mailing list for the GCC project.
Re: pthread_cancel and C++11
- From: Jonathan Wakely <jwakely dot gcc at gmail dot com>
- To: John Saxton <john dot saxton at elementaltechnologies dot com>
- Cc: gcc-help <gcc-help at gcc dot gnu dot org>
- Date: Tue, 11 Aug 2015 15:51:31 +0100
- Subject: Re: pthread_cancel and C++11
- Authentication-results: sourceware.org; auth=none
- References: <CANZ7fQ+1h4Uw8pQ9qN8mhBkRb2d0HUGGWS2u892sBGg_wbmPHg at mail dot gmail dot com>
On 10 August 2015 at 20:01, John Saxton
<john.saxton@elementaltechnologies.com> wrote:
> My organization recently upgraded to C++11 and we're running into a
> problem. We use the POSIX threading library, and in some rare cases
> (pthread_join doesn't exit within a specified timeout, for example),
> we'll kill a thread via pthread_cancel. In C++, thread cancellation is
> implemented via the abi::__forced_unwind exception[1], and in C++03,
> this worked well for us.
>
> In C++11, all destructors are noexcept(true) by default. So if you try
> to cancel a thread that's in a destructor, the abi::__forced_unwind
> exception will escape the destructor, std::terminate() will be
> invoked, and your program dies. I have written a simple program that
> can reproduce this.[2]
Yes, I've been aware of this problem for some time, and don't have any solution.
> What's the best way to solve this problem? I could avoid using
> pthread_cancel, but I was hoping to be able to upgrade to C++11
> without making major changes like that. I could flag all my
> destructors as noexcept(false), but that's tedious and doesn't cover
> third party code like the STL or Boost. I briefly looked into
> std::thread, but it looks like std::thread doesn't have a cancellation
> mechanism. I have reproduced this with a couple of different libstdc++
> versions (4.8.4-2ubuntu1~14.04 being the newest package I have tested
> with). Would upgrading to a newer version fix my problem?
No, the problem is inherent to using forced_unwind in C++11.
> Any other
> ideas? Do I just need to accept the fact that C++11 broke POSIX thread
> cancellation and there's always going to be a chance that
> pthread_cancel will cause my application to terminate?
Yes, or accept that you shouldn't use pthread_cancel in C++11. I don't
have any better suggestions, sorry.