JonY wrote:
On 01/26/2017 10:30 AM, Christian Franke wrote:
After upgrading to gcc test version 6.3.0-1, C++ exception handling is
broken if DLL version of new libstdc++6 is used.
Noted, I'll consider the test version broken for now.
Meantime I examined the testcase with gdb: The DLL exports a bogus
address for the symbol "_ZTIi" ('typeinfo for int'). This results in
this more simple testcase:
#include <typeinfo>
int main() {
const std::type_info & ti = typeid(int); // &_ZTIi from DLL
const char * p = ti.name(); // segfaults because &_ZTIi is bogus.
}
And indeed, the export table of the DLL looks 'interesting'. The offsets
of various exported symbols are apparently all 93980000:
$ objdump -p cygstdc++-6.dll
[5142] +base[5143] c6760 Export RVA
[5143] +base[5144] c6768 Export RVA
[5144] +base[5145] 93980000 Export RVA
...
[5151] +base[5152] 93980000 Export RVA
[5152] +base[5153] 93980000 Export RVA
[5153] +base[5154] 93980000 Export RVA
[5154] +base[5155] 93980000 Export RVA
...
[5167] +base[5168] 93980000 Export RVA
[5168] +base[5169] c6780 Export RVA
...
[Ordinal/Name Pointer] Table
...
[5143] _ZTISt9type_info
[5144] _ZTIa
...
[5151] _ZTIh
[5152] _ZTIi
[5153] _ZTIj
[5154] _ZTIl
...
[5167] _ZTSDs
[5168] _ZTSN10__cxxabiv115__forced_unwindE
$ objdump -p cygstdc++-6.dll | grep -cw 93980000
759
Hmm... bug in the (bin)utils used to build the DLL?
Possibly related:
https://cygwin.com/ml/cygwin/2017-01/msg00375.html
Christian
--
Problem reports: http://cygwin.com/problems.html
FAQ: http://cygwin.com/faq/
Documentation: http://cygwin.com/docs.html
Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple