Segfault on app start using Trusty Beta with packaged PCL

[kubat]

I am trying out the Ubuntu 14.04LTS Beta (Trusty) and have been seeing a strange error.
When using the prebuilt packages by Jochen Sprickerhof, my apps segfault on start if I compile with c++11 enabled.

• This does not occur without c++11 enabled.
• This does not occur when I compile and install from sources.

From within GDB, I see the following trace for the crashes:

`(gdb) run
Program received signal SIGSEGV, Segmentation fault.
0x00007fffed3a1bb0 in boost::math::lanczos::lanczos_initializer<boost::math::lanczos::lanczos17m64, long double>::init::
init() () from /usr/lib/libpcl_features.so.1.7
(gdb) bt
#0 0x00007fffed3a1bb0 in boost::math::lanczos::lanczos_initializer<boost::math::lanczos::lanczos17m64, long double>::in

it::init() () from /usr/lib/libpcl_features.so.1.7
#1 0x00007ffff7dea13a in call_init (l=, argc=argc@entry=1, argv=argv@entry=0x7fffffffdc58, env=env@entry

=0x7fffffffdc68) at dl-init.c:78
#2 0x00007ffff7dea223 in call_init (env=, argv=, argc=, l=)

at dl-init.c:36
#3 _dl_init (main_map=0x7ffff7ffe1c8, argc=1, argv=0x7fffffffdc58, env=0x7fffffffdc68) at dl-init.c:126
#4 0x00007ffff7ddb30a in _dl_start_user () from /lib64/ld-linux-x86-64.so.2
#5 0x0000000000000001 in ?? ()
#6 0x00007fffffffdfda in ?? ()
#7 0x0000000000000000 in ?? ()`

Jochen was kind enough to help me build my own .debs from sources, and these seem to exhibit the same problem.

Details of my setup:

• Ubuntu 14.04LTS Beta
• Running on a VMWare VM on a Macbook Pro (i7)
• Using libboost-math1.54-dev version 1.54.0-4ununtu3 (installed deb)

[Breakthru]

I get the same error on a call generated from the sample_consensus module.
In my case, it happens when CMAKE_BUILD_TYPE is set to Debug (and not on Relase)
Does not happen with a build from source on the same machine.
I'm compiling with C++11 using the flag -std=c++0x
Using Ubuntu 14.04 with libpcl-1.7-all from packages, and g++ (Ubuntu 4.8.2-19ubuntu1) 4.8.2

gdb backtrace:
Program received signal SIGSEGV, Segmentation fault.
0x00007ffff1e55c50 in boost::math::lanczos::lanczos_initializer<boost::math::lanczos::lanczos17m64, long double>::init::init() () from /usr/lib/libpcl_sample_consensus.so.1.7
(gdb) bt
#0 0x00007ffff1e55c50 in boost::math::lanczos::lanczos_initializer<boost::math::lanczos::lanczos17m64, long double>::init::init() () from /usr/lib/libpcl_sample_consensus.so.1.7
#1 0x00007ffff1e2d6de in ?? () from /usr/lib/libpcl_sample_consensus.so.1.7

[taketwo]

This is a know problem, see this or this. In short, you should not mix libraries compiled with different versions of C++ standard. So either build everything from source with C++11 support by yourself, or use system libraries and limit yourself to the old standard. That being said, in many cases you could get lucky and mixing will work just fine.

[jcar87]

As taketwo has mentioned, this is a known problem with Boost, when mixing binaries that include boost/math/special_functions/lanczos.hpp , and one of the binaries is built with c++11 and the other isn't.

If you are building your own project with c++11, you may be better off building PCL with c++11 mode as well (see this patch )

If this isn't possible, I'd suggest using boost 1.60, which apparently fixes this issue. But you would also have to re build PCL (c++11 or not), with that version of boost.

[liuzhenboo]

This is a know problem, see this or this. In short, you should not mix libraries compiled with different versions of C++ standard. So either build everything from source with C++11 support by yourself, or use system libraries and limit yourself to the old standard. That being said, in many cases you could get lucky and mixing will work just fine.

我解决了：catkin_make -DCMAKE_BUILD_TYPE=Release