FTBFS on i386: dbaccess_RowSetClones unit test segfaults

I wonder if that could be linked to bug #1699772 ?

Full stack trace:

Thread 1 "cppunittester" received signal SIGSEGV, Segmentation fault.
0xab9cc965 in _expand_stack_to(unsigned char*) () from /usr/lib/jvm/java-8-openjdk-i386/jre/lib/i386/server/libjvm.so
(gdb) bt
#0 0xab9cc965 in _expand_stack_to(unsigned char*) () at /usr/lib/jvm/java-8-openjdk-i386/jre/lib/i386/server/libjvm.so
#1 0xab9cf174 in os::Linux::manually_expand_stack(JavaThread*, unsigned char*) ()
    at /usr/lib/jvm/java-8-openjdk-i386/jre/lib/i386/server/libjvm.so
#2 0xab9d96b8 in os::create_main_thread(JavaThread*) () at /usr/lib/jvm/java-8-openjdk-i386/jre/lib/i386/server/libjvm.so
#3 0xabb1bece in Threads::create_vm(JavaVMInitArgs*, bool*) () at /usr/lib/jvm/java-8-openjdk-i386/jre/lib/i386/server/libjvm.so
#4 0xab7c8635 in JNI_CreateJavaVM () at /usr/lib/jvm/java-8-openjdk-i386/jre/lib/i386/server/libjvm.so
#5 0xb27a57d1 in jfw_plugin_startJavaVirtualMachine(JavaInfo const*, JavaVMOption const*, long, JavaVM_**, JNIEnv_**) (pInfo=<optimized out>, arOptions=<optimized out>, cOptions=<optimized out>, ppVm=<optimized out>, ppEnv=<optimized out>)
    at ./jvmfwk/plugins/sunmajor/pluginlib/sunjavaplugin.cxx:845
#6 0xb27b5e94 in jfw_startVM(JavaInfo const*, JavaVMOption*, long, JavaVM_**, JNIEnv_**) (pInfo=<optimized out>, arOptions=<optimized out>, cOptions=<optimized out>, ppVM=<optimized out>, ppEnv=<optimized out>) at ./jvmfwk/source/framework.cxx:340
#7 0xad1f7ad0 in stoc_javavm::JavaVirtualMachine::getJavaVM(com::sun::star::uno::Sequence<signed char> const&) (this=0x80b15b28, rProcessId=uno::Sequence of length 17 = {...}) at ./stoc/source/javavm/javavm.cxx:759
#8 0xaf40cc97 in gcc3::callVirtualMethod(void*, long, void*, _typelib_TypeDescription*, bool, long*, long) (pAdjustedThisPtr=0x80b15b58, nVtableIndex=3, pRegisterReturn=0xbfff21f0, pReturnTypeDescr=0x8009a5f8, bSimpleReturn=false, pStackLongs=0xbfff2210, nStackLongs=3)
    at /home/ubuntu/build/libreoffice-5.3.3/bridges/source/cpp_uno/gcc3_linux_intel/callvirtualmethod.cxx:102
#9 0xaf40c249 in (anonymous namespace)::cpp_call(bridges::cpp_uno::shared::UnoInterfaceProxy*, bridges::cpp_uno::shared::VtableSlot, typelib_TypeDescriptionReference*, sal_Int32, typelib_MethodParameter*, void*, void**, uno_Any**) (pThis=pThis@entry=0x80b1f058, aVtableSlot=..., pReturnTypeRef=<optimized out>, nParams=<optimized out>, pParams=<optimized out>, pUnoReturn=<optimized out>, pUnoArgs=<optimized out>, ppUnoExc=<optimized out>) at ./bridges/source/cpp_uno/gcc3_linux_intel/uno2cpp.cxx:161
#10 0xaf40c85b in bridges::cpp_uno::shared::unoInterfaceProxyDispatch(_uno_Interface*, _typelib_TypeDescription const*, void*, void**, _uno_Any**) (pUnoI=0x80b1f058, pMemberDescr=0x80b14310, pReturn=0xada109c0, pArgs=0xada109e0, ppException=0xada10a3c)
    at ./bridges/source/cpp_uno/gcc3_linux_intel/uno2cpp.cxx:360
#11 0xada13f11 in s_dispatcher_v(va_list*) (pParam=0xada10988) at ./cppu/source/helper/purpenv/helper_purpenv_Proxy.cxx:384
#12 0xb77e2c9c in s_environment_invoke_v(uno_Environment*, uno_Environment*, uno_EnvCallee*, va_list*) (pCurrEnv=0x0, pTargetEnv=<optimized out>, pCallee=0xada13ee0 <s_dispatcher_v(va_list*)>, pParam=0xada10988) at ./cppu/source/uno/EnvStack.cxx:293
#13 0xa...

If that is indeed bug #1699772, then updating the kernel on the launchpad builders should fix it.

I just did a (failed) build in a PPA, and the kernel version was:
    Kernel reported to sbuild: 4.4.0-81-generic #104-Ubuntu SMP Wed Jun 14 08:17:06 UTC 2017 i686

Supposedly kernel 4.4.0-83.106 has the fix.

I am seeing a different unit test failure when building i386 snap packages for libreoffice, but I am suspecting this is caused by the same issue:

   [build CUT] dbaccess_hsqldb_test
   Segmentation fault (core dumped)

(attaching full build log)

An i386 build of libreoffice 5.3.4 in a PPA (with -proposed enabled) still fails on the same unit test. In fact, two unit tests fail:

    dbaccess_hsqldb_test.test
    dbaccess_RowSetClones.test

Which is consistent with the snap build, except that the snap build fails on the first one and bails out, whereas the PPA build appears to be running more tests before it bails out.

And both builds (snap and PPA) were performed with kernel 4.4.0-83-generic #106.

It’s been reported in debian that libreoffice writer is crashing at startup with java enabled, on 32 bits only with their latest kernel update. This smells like the same issue, and one that is affecting the actual application, not just the unit tests. See https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=865303 for details.

This bug was fixed in the package libreoffice - 1:5.3.4-0ubuntu1

---------------
libreoffice (1:5.3.4-0ubuntu1) artful; urgency=medium

  * New upstream release: 5.3.4
  * debian/rules: temporarily ignore unit test failures on i386 (LP: #1700692)

 -- Olivier Tilloy <email address hidden> Thu, 13 Jul 2017 18:30:32 +0200