Why eh_frame sections in libgcc.a?

[Andrew Haley <aph at redhat dot com>]

Stefan Bylund writes:
 > When building GCC 3.2 for powerpc-eabi, ".eh_frame" exception handling sections
 > are generated in libgcc.a for the following objects:
 > 
 > _divdi3.o
 > _moddi3.o
 > _udivdi3.o
 > _umoddi3.o
 > _udiv_w_sdiv.o
 > _udivmoddi4.o
 > unwind-dw2.o
 > unwind-dw2-fde.o
 > 
 > I can understand why such sections are generated for unwind-dw2.o and
 > unwind-dw2-fde.o which are used by libstdc++.a and libsupc++.a but why are they
 > generated for the other integer division related objects?

Because Java throws an exception on division by zero.

 > Is there support for throwing C++ exceptions when dividing by zero?

You'll need -fnon-call-exceptions to make this work.  You'll also need
some code to trap the signal and trun it into an exception; examples
are in libgcj.  This does not work on all platforms.

 > What happens with such support in a plain C application?

It's irrelevant unless an application catches the signal and throws an
exception.  Plain old C can make a signal handler and longjmp() out of
it.

 > Are these ".eh_frame" sections in libgcc.a unused and harmless when
 > compiling a plain C application and linking it with libgcc.a?

Yes.

Andrew.