1.4 on Linux: segfault

[Linzer]

Hi,
I have compiled BfW 1.4 on Mandriva Linux 2008, 32-bit. It segfaults systematically on startup. Here is the backtrace:

Code: Select all

#0  0x084f0be2 in std::locale::operator= ()
#1  0x084ea39d in std::ios_base::_M_init ()
#2  0x084e8f2d in std::basic_ios<char, std::char_traits<char> >::init ()
#3  0x083ce7ca in __static_initialization_and_destruction_0 ()
    at /usr/lib/gcc/i586-mandriva-linux-gnu/4.2.2/../../../../include/c++/4.2.2/bits/locale_facets.tcc:2557
#4  0x0852bec5 in __do_global_ctors_aux ()
#5  0x080733ac in _init ()
#6  0x0852be69 in __libc_csu_init ()
#7  0xb7999f31 in __libc_start_main () from /lib/i686/libc.so.6
#8  0x08074971 in _start ()

Any ideas?
Linzer

[Mist]

It looks like Mandriva specific glibc segfault, does official package behave like this?

[Linzer]

I tried the official package, but it is only provided for the development version of the distribution (Cooker), and I am running a stable version (2008). It requires a more recent version of boost than the standard one on 2008, so I uninstalled boost and try to compile the newer one. Now the fun starts: while trying to compile boost, I get another segfault! And the additional fun is, if I enable debug symbols in that process, then the segfault magically disappears.
Either it's the same glibc bug, or another, boost-specific bug...
My next step is to trying to complete the boost install with debug symbols and see if wesnoth survives. To be continued...

Edit: That seems to have worked. To recap, the official build with an additional libboost_iostreams-mt.so.1.34.1 (home-built) seems to run okay.