Re: exceptions in gnu c++

[Paul E Condon <pecondon@mesanetworks.net>]

On Sun, Nov 05, 2006 at 10:51:15PM +0100, Christian Holm Christensen wrote:
> Hi Paul,
> 
> On Sun, 2006-11-05 at 14:29 -0700, Paul E Condon wrote:
> > I've written a simulation program in c++. It works for
> > small problems of its class, but fails during initialization
> > on large problems. I suspect I need more RAM, but would
> > like confirmation, so ...
> > 
> > I changed the main program to include a everything in a
> > try block with a catch block, etc. But it still dies in
> > the same way. Where can I find an example of exception
> > handling code that works with gcc 3.3? 
> 
> Can you switch to 3.4+? 3.3 is buggy. 
> 
> > My code is:
> > using namespace std;
> > 
> > #include "Sys.hh"
> > #include "nmlib.hh"
> > #include "rgen2.hh"
> > #include "nmsim.hh"
> > 
> > int main( int argc, char* argv[] ){
> >   cout<<"Starting nmmain..."<<endl;
> >   try {
> >     NMsim X;                 // instantiate the simulation object
> >     X.nminit( argc, argv ); // read in parameters and generate starting conditions
> >     X.nmcore( );         // do a simulation run
> >   } catch (const exception caught) {
> 
> You should catch a reference to an exception object - not a real object.
> The reason is, that you'd only get a copy, and that may not be done
> right.  So you should have 
> 
>     } catch (const exception& caught) { 
> 
> >     cout << "caught exception: "<<caught.what() <<"!!!"<< endl;
> 
> Perhaps you should return a non-null value here.  Also, perhaps you want
> to dump error messages to `std::cerr' and debug messages to
> `std::clog'. 
> 
> >   }
> >   cout<<" After catch."<<endl;
> >   return 0;
> > };
> > 
> > NMsim::NMsim(){};
> > 
> > </end>
> > 
> > The output is:
> > gq:/da4/wdl/H$ nmmain --steps 200 --stpsiz 200 --kernparm 1024. vaa.a
> > Starting nmmain...
> > init set to 255
> >  done with tbl/abn/sid preparation
> >  start dispersal calc with rx=5418
> >  row: 21100     128    Killed
> > gq:/da4/wdl/H$
> > </end>
> > 
> > Notice the 'debug' statements being sent to cout.
> > 'Killed' comes from some gcc library, not from my code, 
> > especially not from my catch block. 
> 
> I think the reference will fix it.  If not, try running the program
> through GDB and set a break-point in `std::terminate' (the function
> called for un-caught exceptions) and then back-trace. 
> 
> > The detailed
> > syntax that I'm attempting to use is taken from
> > Josuttis, p. 15. 
> 
> Perhaps a URL would be good? 
> 
> > He footnotes it with a caveat that
> > it is not in the Standard, only proposed. Looks like
> > something in the c++ library is catching whatever
> > is being thrown (if anything) and killing nmmain
> > before control gets to my catch and/or return stmnts.
> 
> Ye, that would be `std::terminate'.  However, the caveat is probably no
> longer valid.  The `std::terminate' function will be called if your
> catch blocks does not match the exception thrown.   The C++ library only
> throws exceptions that inherit from std::exception, so that should be
> caught.  However, if some code throws a std::string or something, that
> will not be caught by  your catch block.  If you want to catch all
> exceptions, no matter the type, you could append _another_ catch block
> like 
> 
>   catch (...) {
>     std::cerr << "Caught unknown exception" << std::endl;
>   }
> 
> > suggestions? (like maybe a more appropriate list to ask?)
> 
> A list dedicated to C++ questions would be an obvious choice :-) 
> 
> Yours,

I have a new version of my program and I have switched to g++-3.4
The newer version addresses issues about the actual name of the
exception:
#include "Sys.hh"
#include "nmlib.hh"
#include "rgen2.hh"
#include "nmsim.hh"

int main( int argc, char* argv[] ){
  clog<<"Starting nmmain..."<<endl;
  try {
    NMsim X;                 // instantiate the simulation object
    X.nminit( argc, argv ); // read in parameters and generate starting conditions
    X.nmcore( );         // do a simulation run
  } 
  catch (...) {
    clog << "caught ...!!!"<< endl;
  }
  clog<<" After catch."<<endl;
  return 0;
};
</end>

The output still shows no sign that the catch block is entered.
Still the same 'Killed' word from somewhere.
Should I be able to put essentially all of a program inside
a try block like I'm doing here? It is hard to believe that this
is an undiscovered bug in gnu c++. What might be missing from
my code? I've never done try/catch before so don't hesitate
to suggest the 'obvious'.

-- 
Paul E Condon           
pecondon@mesanetworks.net