I omitted one important detail:
$ g++ -v
Using built-in specs.
Target: powerpc-ibm-aix5.3.0.0
Configured with: ../configure --prefix=/usr/local/gcc-4.2.2
--enable-languages=c,c++ --with-as=/usr/bin/as --with-ld=/usr/bin/ld
--disable-nls --enable-threads --enable-version-specific-runtime-libs
--host=powerpc-ibm-aix5.3.0.0
Thread model: aix
gcc version 4.2.2
Also, trying to use the at_exit mechanism for destruction does not work:
$ g++ -fuse-cxa-atexit t2.cpp
ld: 0711-317 ERROR: Undefined symbol: __dso_handle
ld: 0711-317 ERROR: Undefined symbol: .__cxa_atexit
Michael
Veksler/Haifa/IBM
To
03/03/08 03:59 gcc-help@xxxxxxxxxxx
cc
Subject
AIX/g++: Preventing early
destruction of exception handling
The following trivial code crashes on AIX-5.3 + gcc-4.2.2 (same with
gcc-3.4.3):
$ cat t.cpp
struct A { ~A() { try {throw 1;} catch(...) {} } };
A a;
int main() { }
$ g++ -g t.cpp
$ ./a.out
Abort
It looks like the libgcc_s.a(shr.o) library is destroyed (along with
exception supporting data structures) before the destructor of A has been
called. This is done by a function named _GLOBAL__FD_shr_o which seems to
be responsible for destruction of all global objects in libgcc.
When the destructor of A is called (by _GLOBAL__FD_a_out), exception
handling mechanism is already dead.
Is there anything constructive I can do to overcome this hurdle?
Should I open a PR (unless there is one, already open).
As far as I can tell, my code is well specified and defined from the ISO-C+
+ point of view, and is accepted on gcc+Linux and xlC+AIX. Only gcc+AIX
happens to problematic.
Thanks
Michael
