bp::extend::handler in V2?

[db4]

For V1 I used the following trick to kill a child Linux process when its parent dies:

struct ExecHandler : bp::extend::handler
{
    template<typename Executor>
    void on_exec_setup (Executor &) const
    {
        (void)prctl(PR_SET_PDEATHSIG, SIGTERM);
    }
};
...
    bp::child p(
        ExecHandler(),
        ...
    );

Is there anything similar to bp::extend::handler in V2?

[klemens-morgenstern]

Similar code, just without the base-class:

struct ExecHandler 
{

    error_code on_exec_setup(posix::default_launcher & /*launcher*/, const filesystem::path & /* exe*/, const char * const * /* argv*/)
   {
              (void)prctl(PR_SET_PDEATHSIG, SIGTERM);
   }

};

process::v2::process p{..., ExecHandler{}};

[db4]

@klemens-morgenstern, thanks a lot! If you don't mind, I have a similar question regarding Windows. I used to achieve the same effect of killing a process when its parent dies with

            bp::group g;
            bp::child p(
                ...
                g,
                io_service
            );
#ifdef _WIN32
            namespace bpd = bp::detail;
            bpd::api::workaround::JOBOBJECT_EXTENDED_LIMIT_INFORMATION_ info;
            auto h = g.native_handle();
            if (!bpd::api::workaround::query_information_job_object(
                h,
                bpd::api::workaround::JobObjectExtendedLimitInformation_,
                static_cast<void*>(&info),
                sizeof(info),
                nullptr)) {
                bpd::throw_last_error("QueryInformationJobObject() failed");
            }
            info.BasicLimitInformation.LimitFlags |= JOB_OBJECT_LIMIT_KILL_ON_JOB_CLOSE;
            if (!bpd::api::workaround::set_information_job_object(
                h,
                bpd::api::workaround::JobObjectExtendedLimitInformation_,
                static_cast<void*>(&info),
                sizeof(info))) {
                bpd::throw_last_error("SetInformationJobObject() failed");
            }
#endif

How to do this with V2? With bp::group no longer existing I don't see how to include a process into a group and get the group's handle.

[klemens-morgenstern]

I did not add the group to v2 because there are too many differences between the windows & linux implementation. And it's simple enough, you just got a bit of boilerplate re HANDLEs and error handling.

The whole api::workaround stuff is also moot, because the

You would need to use the job object directly:

#include <windows.h> 
#include <boost/process/windows/creation_flag.hpp>

auto h = ::CreateJobObjectW(nullptr, nullptr); // don't forget the CloseHandle


bp::process proc(..., boost::process::windows::create_breakaway_from_job);

AssignProcessToJobObject(h, proc.native_handle()); // error handling...

// now do with the job object `h` what you have above

[klemens-morgenstern]

I'm assuming this does it, please reopen or create a new issue if you run into problems.

[db4]

@klemens-morgenstern, let me return to this issue. Your solution

#include <windows.h> 
#include <boost/process/windows/creation_flag.hpp>

auto h = ::CreateJobObjectW(nullptr, nullptr); // don't forget the CloseHandle


bp::process proc(..., boost::process::windows::create_breakaway_from_job);

AssignProcessToJobObject(h, proc.native_handle()); // error handling...

// now do with the job object `h` what you have above

seems to have a flaw. If the child process manages to create a grandchild process before AssignProcessToJobObject is called, the grandchild won't be killed be the OS on the main process crash. V1 allowed to include the child process into the main process' group before it runs; for v2 it's not possible. Another option is to create the child process suspended, as recommended here: https://devblogs.microsoft.com/oldnewthing/20131209-00/?p=2433, but to resume it one have to know the main thread handle, that cannot be fetched from bp::process object. So currently there is no solution that works with v2.