static ObjectFactory access in OmeroCpp may need a mutex

[jamoore]

From Bernhard:

A colleague of mine has tried to create clients in parallel
threads, this caused a segfault while accessing the objectfactories
map. Could it be that you need to mutex the objectFactories method? If
one thread added some keys already, while another is accessing the
partial filled map leads to missing keys. I could be something
different though, as I know that you have a mutex in the client code
above the access to the objectFactories method.

[jmoore]

From Bernhard:

I just read a thread on comp.lang.c++ about threads and singleton
initialization. They suggested the following:

As Bart van Ingen Schenau and James Kanze have said, it's thread-safe
as is, but there's the static initialization order problem. Generally,
you solve this problem either with library support ala pthread_once,
or you make your singleton instance be a function local static to
"initialize on demand" to solve the static initialization order
problem, and then you make a global whose sole purpose is to
initialize the singleton before main(), and thus presumably before
there are threads. (If there are nontrivial threads before main, you
cannot solve this problem with mutexes alone in C++. You would need
the power of pthread_once or equivalent.)

Ex:

//hpp
class singleton
{
public:
    static singleton& getSingleton();

};

//cpp
singleton& singleton::getSingleton()
{   static singleton * x = new singleton;
    return *x;
}

namespace { bool forceInit = (singleton::getSingleton(), true); }

See ​http://groups.google.com/group/comp.lang.c++/browse_frm/thread/9007b1aacad2022e?hl=en
for details.

[jmoore]

r5498 refactors the RTypes object factory registration. I never reproduced the issue, so let me know if it persists.