strange segfault involving io::readWkt, algorithm::volume and CGAL::Nef_polyhedron_3

[jjpepper]

This simple sample program causes a segfault, but commenting out the line which declares a CGAL::Nef_polyhedron_3 in foo() and it all works and outputs "block volume = 1".

My system is RHEL6 building with gcc4.4 and using SFCGAL 1.3.0 and CGAL 4.9. SFCGAL and CGAL built from source. I'm compiling with -frounding-math -g -O0 -std=c++0x

I put the Nef_polyhedron_3 into a separate function to be sure it wasn't a stack issue, but the same problem happens if the call to foo() is replaced with the body of foo(). I get an empty stack with CGAL::internal::intersection is the current stack location. Stack #1 is 0x00 (see output below).

#include <SFCGAL/io/wkt.h>
#include <SFCGAL/PolyhedralSurface.h>
#include <SFCGAL/algorithm/volume.h>
#include <SFCGAL/Solid.h>
#include <CGAL/Nef_polyhedron_3.h>

const std::string block0( "POLYHEDRALSURFACE Z (((0 0 0, 0 1 0, 1 0 0, 0 0 0 )), "
                                        "((0 0 0, 1 0 0, 0 0 1, 0 0 0 )), "
                                        "((0 0 0, 0 0 1, 0 1 0, 0 0 0 )), "
                                        "((1 0 0, 0 1 0, 0 0 1, 1 0 0 )) )" );

void foo()
{
    CGAL::Nef_polyhedron_3<SFCGAL::Kernel> n;
}

int main()
{
    //::SFCGAL::Solid solid; // This instead of next 2 lines works fine
    ::std::auto_ptr< ::SFCGAL::Geometry > geometry0( ::SFCGAL::io::readWkt( block0 ) );
    ::SFCGAL::Solid solid( geometry0->as< ::SFCGAL::PolyhedralSurface >() );
    auto vol = ::SFCGAL::algorithm::volume( solid );
    std::cerr << "block volume = " << vol << "\n";
    foo();
    return 0;
}

gdb output:

(gdb) run
Starting program: /home/jesse.j.pepper/tmp/volume 
[Thread debugging using libthread_db enabled]

Program received signal SIGSEGV, Segmentation fault.
0x00007ffff751d96e in CGAL::Intersection_traits<CGAL::Simple_cartesian<CGAL::Interval_nt<false> >, CGAL::Simple_cartesian<CGAL::Interval_nt<false> >::Triangle_3, CGAL::Simple_cartesian<CGAL::Interval_nt<false> >::Triangle_3>::result_type CGAL::internal::intersection<CGAL::Simple_cartesian<CGAL::Interval_nt<false> > >(CGAL::Simple_cartesian<CGAL::Interval_nt<false> >::Triangle_3 const&, CGAL::Simple_cartesian<CGAL::Interval_nt<false> >::Triangle_3 const&, CGAL::Simple_cartesian<CGAL::Interval_nt<false> > const&) ()
   from /home/jesse.j.pepper/git/builder-code/work/rhel6_64-gcc4_4-sfcgal-1_3_0/binaries/lib64/libSFCGAL.so.1
Missing separate debuginfos, use: debuginfo-install glibc-2.12-1.192.el6.x86_64 libgcc-4.4.7-17.el6.x86_64 libstdc++-4.4.7-17.el6.x86_64
(gdb) bt
#0  0x00007ffff751d96e in CGAL::Intersection_traits<CGAL::Simple_cartesian<CGAL::Interval_nt<false> >, CGAL::Simple_cartesian<CGAL::Interval_nt<false> >::Triangle_3, CGAL::Simple_cartesian<CGAL::Interval_nt<false> >::Triangle_3>::result_type CGAL::internal::intersection<CGAL::Simple_cartesian<CGAL::Interval_nt<false> > >(CGAL::Simple_cartesian<CGAL::Interval_nt<false> >::Triangle_3 const&, CGAL::Simple_cartesian<CGAL::Interval_nt<false> >::Triangle_3 const&, CGAL::Simple_cartesian<CGAL::Interval_nt<false> > const&)
    () from /home/jesse.j.pepper/git/builder-code/work/rhel6_64-gcc4_4-sfcgal-1_3_0/binaries/lib64/libSFCGAL.so.1
#1  0x0000000000000000 in ?? ()
(gdb) 

Interestingly, running in valgrind shows a CGAL::Assertion_exception being printed out which is not present when running on the command line or in gdb, despite the fact that I am definitely compiling with -frounding-math.

Valgrind output:

[Thu May 25, 17:48:18    ~/tmp  jesse.j.pepper@peps (0)] 
$ valgrind ./volume
==29016== Memcheck, a memory error detector
==29016== Copyright (C) 2002-2012, and GNU GPL'd, by Julian Seward et al.
==29016== Using Valgrind-3.8.1 and LibVEX; rerun with -h for copyright info
==29016== Command: ./volume
==29016== 
terminate called after throwing an instance of 'CGAL::Assertion_exception'
  what():  CGAL ERROR: assertion violation!
Expr: -CGAL_IA_MUL(-1.1, 10.1) != CGAL_IA_MUL(1.1, 10.1)
File: /home/jesse.j.pepper/git/mas/builder/3rd/cgal/include/CGAL/Interval_nt.h
Line: 209
Explanation: Wrong rounding: did you forget the  -frounding-math  option if you use GCC (or  -fp-model strict  for Intel)?
==29016== 
==29016== HEAP SUMMARY:
==29016==     in use at exit: 21,095 bytes in 91 blocks
==29016==   total heap usage: 110 allocs, 19 frees, 22,865 bytes allocated
==29016== 
==29016== LEAK SUMMARY:
==29016==    definitely lost: 0 bytes in 0 blocks
==29016==    indirectly lost: 0 bytes in 0 blocks
==29016==      possibly lost: 1,023 bytes in 7 blocks
==29016==    still reachable: 20,072 bytes in 84 blocks
==29016==         suppressed: 0 bytes in 0 blocks
==29016== Rerun with --leak-check=full to see details of leaked memory
==29016== 
==29016== For counts of detected and suppressed errors, rerun with: -v
==29016== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 6 from 6)
Aborted (core dumped)
=================================================================================
[Thu May 25, 17:48:21    ~/tmp  jesse.j.pepper@peps (0)] 
$ 

[jjpepper]

I tried this with gcc6.3.1 20170216 and still get the same behaviour.

[jjpepper]

Is anyone able to try this on a different platform?

[mhugo]

I reproduce with gcc 5.4.0 on ubuntu 16.04 with CGAL 4.7
But I have really no clue about what happens ... :/

It might be a CGAL issue. It would be interesting to try a similar code without SFCGAL. Something like a call to CGAL::intersection with and without Nef_Polyhedron ...

btw, CGAL is known to have problem with valgrind. Maybe try this https://github.com/SolidCode/pyscad/blob/master/patches/CGAL-Valgrind-Enable-Hack.patch ?