Dealii: Can we make Subscriptor::list_subscribers work with threads?

[jppelteret]

I'm trying to debug an issue related to a manifold that's somehow associated to an object that's gone out of scope. Since I built deal.II with threads enabled, there's no mechanism to work out which object this manifold is stubbornly attached to. Is there any technical reason as to why the list_subscribers() function of the Subscriptor class cannot be made to work with threads? I was thinking about changing its implementation from

void Subscriptor::list_subscribers () const
{
#ifndef DEAL_II_WITH_THREADS
  for (map_iterator it = counter_map.begin();
       it != counter_map.end(); ++it)
    deallog << it->second << '/'
            << counter << " subscriptions from \""
            << it->first << '\"' << std::endl;
#else
  deallog << "No subscriber listing with multithreading" << std::endl;
#endif
}

to

void Subscriptor::list_subscribers () const
{
#ifdef DEAL_II_WITH_THREADS
  static std::mutex mutex;
  std::lock_guard<std::mutex> lock(mutex);
#endif

  for (map_iterator it = counter_map.begin();
       it != counter_map.end(); ++it)
    deallog << it->second << '/'
            << counter << " subscriptions from \""
            << it->first << '\"' << std::endl;
}

or, even better, implement the above in a more general function

template <typename StreamType>
void
Subscriptor::list_subscribers(StreamType &stream) const

?

[drwells]

I do not see any reason to disable this when threads are present; I am not aware of any place in the library where subscription is performance critical, so adding a lock or atomics should not have a noticeable effect.

The original commit for this feature (819cf4f42840c28eaebac77939e887f2333ab9aa) adds it without comment; I suspect the original author just made the (justifiable) conservative choice that its not clear how to do things with threads safely so this optional feature should be disabled.

[masterleinad]

The problem here is not reading, but writing. Your std::mutex must in fact live in Subscriptor and be also locked (std::lock_guard) when counter_map is modified in subscribe and unsubscribe.

[jppelteret]

Hmm... I see. Thanks a lot for the info and tips, @drwells and @masterleinad. I'm going to play around with this a bit more and I'll report back when I've done so...