Re: C++ Plugins and virtual destructors.

[David Fang]

> > If the vtable is functioning for normal virtual functions, why wouldn't it 
> > also work for a virtual destructor?  Destructors are not magic; they are 
> > merely functions that get called just before memory is released.  RTTI, a 
> > common culprit in dlopen issues, doesn't get involved here.
>
> Right.  A more important question to ask is if static 
> constructors/destructors get properly executed, including the framework 
> necessary to support C++ exceptions.  The C++ exception issue is significant

Hi,
	My experience with C++/plugins verifies that static constructors 
and destructors are invoked upon plug-in loading -- I often use static 
ctors/dtors to perform additional registering of functions with internal 
data structures upon dlopening.  (compilers used: g++-3.3, 3.4, 4.0, 4.2
on FreeBSD, linux, darwin).  I rely on static ctors/dtors to do all work 
related to resource management.

> since usually the framework is set up by the original program and if the 
> plugin does not properly mesh into it then there is a problem.  There may 
> also be a problem if the program is written and linked as a C program but the 
> plugin is C++.

It shouldn't be too hard to set up an experiment or test case using C++ 
exceptions.  I don't have an answer for this since an exception that 
escapes past main() usually results in an abort trap, skipping all 
exit-time cleanup code.  Plugins are cleanly unloaded when a program exits 
normally. I don't know how to catch an exception outside of main, i.e. 
one thrown and uncaught during statics.

I'm not sure a C++ plug-in would work in C, but if you had a function for 
cleaning up resources (something likely to appear in a dtor), you could
register a cleanup function with atexit().

Fang


David Fang
Computer Systems Laboratory
Electrical & Computer Engineering
Cornell University
http://www.csl.cornell.edu/~fang/
        -- (2400 baud? Netscape 3.0?? lynx??? No problem!)