More C++ Exceptions Issues

[ian at airs.com (Ian Lance Taylor)]

   From: "Rosimildo DaSilva" <rdasilva at connecttel.com>
   Date: Sun, 2 May 1999 13:27:37 -0500

   One of the problems with C++ working a very bizarre manner on RTEMS,
   is the fact that I was compiling the BSP=pc386 with the options:
   ( Very small stack size is another problem ).

	 -O4 -fomit-frame-pointer

   this causes C++ code to cause all kind of problems.
   I have stepped in through the code and GDB shows garbage values
   for some parameters, mainly in calls to C++ ctors.

This is expected behaviour.  You can not easily debug code compiled
with -fomit-frame-pointer.  This is documented in the gcc docs:

`-fomit-frame-pointer'
     Don't keep the frame pointer in a register for functions that
     don't need one.  This avoids the instructions to save, set up and
     restore frame pointers; it also makes an extra register available
     in many functions.  *It also makes debugging impossible on some
     machines.*

That fact that you can not use the debugger does not mean that the
code is incorrect.

   Also,  doing this change makes the code to be about 15% smaller.
   I really did not understand this. I really appreciate if someone could
   explain whar is going on.

   Is there any real reason why the flags -O4 -fomit-frame-pointer in the first
   place ?

On a register-starved machine like the i386, -fomit-frame-pointer
normally produces better code because the compiler does not have to
reserve a register for use as a frame pointer.

The change in size you see between -O2 and -O4 may be arising because
at -O3 and above the compiler will automatically inline functions when
possible.  This is a trade off for decreased execution time at the
cost of increased size.  Using -O4 -fno-inline-functions would disable
the automatic inlining.

Ian