audacious2 crashed with SIGSEGV in std::ios_base::Init::~Init()

StacktraceTop:
 ?? ()
 std::ios_base::Init::~Init ()
 ?? ()
 __cxa_finalize (d=0x7ff36edb3220)
 __do_global_dtors_aux ()

I had this issue once in my virtual machine, but I cannot reproduce it any more.

To reproduce:

rm -rf ~/.config/audacious/
audacious
close it

Forwarded to upstream: http://jira.atheme.org/browse/AUD-264

According to upstream this is a bug in libstdc++. Here's the comment from John Lindgren:

"Ubuntu bug reports show crashing in the deconstructor for the C++ class std::ios_base::Init, which is registered with __cxa_atexit and called by __cxa_finalize, sometimes when a C++ plugin is unloaded and sometimes when the program exits. The fact that not unloading any plugins is a workaround suggests to me that the deconstructor cannot handle being called more than once. My guess is that there is a bug in the Ubuntu 10.10 version of libstdc++ (but not in the 10.04 version or in the Debian version that I use).

Since I don't have Ubuntu, I can't try to reproduce this outside of Audacious for you. But I suggest you try compiling a pure C program that uses g_module_open to load and initialize (similarly to Audacious) at least two C++ modules using iostream and then uses g_module_close to unload them again.

Meanwhile, you should be able to use the workaround of not actually unloading the plugins without any ill side effects."

audacious crashes if audacious-plugins used "gcc" as linker. audacious doesn't crash if "g++" is used as linker in audacious-plugins [1].

[1] Adjust the LD value in buildsys.mk.

This bug was fixed in the package audacious-plugins - 2.4.0-0ubuntu3

---------------
audacious-plugins (2.4.0-0ubuntu3) maverick; urgency=low

  * Replace remove-psf.patch by patch from upstream.
  * Use g++ as linker instead of gcc to fix crash on close (LP: #640732).
 -- Benjamin Drung <email address hidden> Sat, 25 Sep 2010 14:41:51 +0200

That does not fix the bug... it is an STL bug... and it was fixed in GCC 4.5.

Man the Ubuntu QA process is disturbingly incompetent.

His "change" does not fix the bug period. It is a bug in GCC, period.

I have to correct my words: It's not a fix, but a workaround. audacious with audacious-plugins 2.4.0-0ubuntu3 doesn't crash any more on exit. The GCC bug is still open.

Did anyone take a look at http://www.cygwin.com/ml/libc-hacker/2010-05/msg00003.html. (It's a well-known CRASH)

Why invalid?

the upstream report talks about "it is ruled that ..." why?
comment #14 has a link to glibc patch, which apparently was never applied upstream?

I don't quite get you, but the bug is here and we are left with a mere workaround now. And regarding comment #14... I dont see any connection with this bug. Do you?

No, but maybe "The rage of gcc" can enlighten us?

I made some tests: Install audacious-plugins 2.4.0-0ubuntu2 and run "rm -rf ~/.config/audacious/ && audacious" and then quit audacious. It should crash. Then add "CXX=g++-4.5" to the dh_auto_configure command and build and install the package. Run "rm -rf ~/.config/audacious/ && audacious", quit audacious and it shouldn't crash.

This shows me that the bug is in g++-4.4, but not in g++-4.5. Therefore I am reopening the task for gcc-4.4.

This version is outdated and no more supported