Re: [gtkmm] gtk-- exceptions - was Re: Gtk-- FAQ: first draft available

[Karl Nelson <kenelson ece ucdavis edu>]

In message <37989D1C.8210454C@ibmoto.com>Todd Dukes writes
>Karl Nelson wrote:
>> 
>[...]
>> Should Gtk-- place a try/catch block arround the calls to the user
>> callbacks that catchs all to an error message or let it
>> crash GTK+ and hopefully fall through as an uncaught exception?
>> 
>> Or alternatively can we write a gtk_signal_exception_cleanup
>> function that can clean up the mess?
>
>I would like to be able to use exceptions in as transparent a method
>as possible. But if that solution causes ugly code with bad 
>performance, it should be avoided. I have the following questions.

Transparent would be to make this work

  try {
    button.pressed();
  }
  catch () {/*...*/}

work.  However, I don't think that will happen.  

>1. What would a gtk_signal_exception_cleanup function do? I am not
>   sure I understand exactly what we need to fix, or what method
>   is used to fix it.

Well, I assume that the user is saying that a signal path which
throws an exception back through a c function path will cause
problems.  But I don't think that you can ever trust that it 
won't seg fault immediately when it hits gtk+ code.  If it
does work the user would have to call a clean up routine imediately
in the catch.  Not a good idea.  (Or gtk+ would have to be exception
aware, also not too good.)

>2. Where would it go (gtk-- code or user code)? 

Depends on the solution.  A try block in the emitter is
internal to gtk--, but then the user would have to supply
a catch function to gtk-- some how.  (See Esa Pulkkinen's
page on reusing catch functions.)

>3. What would it do to the performance of programs that don't
>   use exceptions?

Well, it takes about 1-5% off most programs to have the try blocks
that are not used.  To verify how much it is on your compiler
use this code

----------------------------------------------------------------
#define TRY 1
int j;
void foo()
  { j++; }

main()
  {
   for (int i=0;i<(2<<24);i++)
     {
#if TRY
      try 
        {
#endif
         foo();
#if TRY
        }
      catch (...) {}
#endif
     }
  }
----------------------------------------------------------------

On my machine....

without try: 4.570u 0.010s 0:04.57 100.2%    0+0k 0+0io 109pf+0w
with try:    4.820u 0.010s 0:04.82 100.2%    0+0k 0+0io 109pf+0w

5% penalty for try blocks, in a trivial program. (less in 
larger programs)

--Karl