1.3.6 build error on Fedora 30 (CGAL 4.14, gcc 9.0.1)

[devrimgunduz]

Hi,

I'm getting this while building SFCGAL on Fedora 30. Any hints how I can fix it?

Thanks!

`/var/lib/pgsql/git/pgrpms/rpm/redhat/master/sfcgal/master/SFCGAL-1.3.6/src/Coordinate.cpp: In member function 'CGAL::Lazy_kernel_generic_base<CGAL::Simple_cartesian<boost::multiprecision::numberboost::multiprecision::backends::gmp_rational >, CGAL::Simple_cartesian<CGAL::Interval_nt >, CGAL::Cartesian_converter<CGAL::Simple_cartesian<boost::multiprecision::numberboost::multiprecision::backends::gmp_rational >, CGAL::Simple_cartesian<CGAL::Interval_nt > >, CGAL::Epeck>::FT SFCGAL::RoundVisitor::_roundFT(const FT&) const':
/var/lib/pgsql/git/pgrpms/rpm/redhat/master/sfcgal/master/SFCGAL-1.3.6/src/Coordinate.cpp:276:72: error: no matching function for call to 'round(boost::enable_if_c<true, boost::multiprecision::detail::expression<boost::multiprecision::detail::multiply_immediates, boost::multiprecision::numberboost::multiprecision::backends::gmp_rational, long int, void, void> >::type)'
276 | SFCGAL::round( v.exact() * _scaleFactor ),
| ^
In file included from /var/lib/pgsql/git/pgrpms/rpm/redhat/master/sfcgal/master/SFCGAL-1.3.6/include/SFCGAL/Coordinate.h:31,
from /var/lib/pgsql/git/pgrpms/rpm/redhat/master/sfcgal/master/SFCGAL-1.3.6/src/Coordinate.cpp:21:
/var/lib/pgsql/git/pgrpms/rpm/redhat/master/sfcgal/master/SFCGAL-1.3.6/include/SFCGAL/numeric.h:50:15: note: candidate: 'double SFCGAL::round(const double&)'
50 | inline double round( const double& v )
| ^~~~~
/var/lib/pgsql/git/pgrpms/rpm/redhat/master/sfcgal/master/SFCGAL-1.3.6/include/SFCGAL/numeric.h:50:36: note: no known conversion for argument 1 from 'boost::enable_if_c<true, boost::multiprecision::detail::expression<boost::multiprecision::detail::multiply_immediates, boost::multiprecision::numberboost::multiprecision::backends::gmp_rational, long int, void, void> >::type' {aka 'boost::multiprecision::detail::expression<boost::multiprecision::detail::multiply_immediates, boost::multiprecision::numberboost::multiprecision::backends::gmp_rational, long int, void, void>'} to 'const double&'
50 | inline double round( const double& v )
| ~~~~~~~~~~~~~~^
/var/lib/pgsql/git/pgrpms/rpm/redhat/master/sfcgal/master/SFCGAL-1.3.6/include/SFCGAL/numeric.h:86:23: note: candidate: 'CGAL::Gmpz SFCGAL::round(const CGAL::Gmpq&)'
86 | SFCGAL_API CGAL::Gmpz round( const CGAL::Gmpq& v ) ;
| ^~~~~
/var/lib/pgsql/git/pgrpms/rpm/redhat/master/sfcgal/master/SFCGAL-1.3.6/include/SFCGAL/numeric.h:86:48: note: no known conversion for argument 1 from 'boost::enable_if_c<true, boost::multiprecision::detail::expression<boost::multiprecision::detail::multiply_immediates, boost::multiprecision::numberboost::multiprecision::backends::gmp_rational, long int, void, void> >::type' {aka 'boost::multiprecision::detail::expression<boost::multiprecision::detail::multiply_immediates, boost::multiprecision::numberboost::multiprecision::backends::gmp_rational, long int, void, void>'} to 'const CGAL::Gmpq&'
86 | SFCGAL_API CGAL::Gmpz round( const CGAL::Gmpq& v ) ;
| ~~~~~~~~~~~~~~~~~~^
make[3]: *** [src/CMakeFiles/SFCGAL.dir/build.make:66: src/CMakeFiles/SFCGAL.dir/Coordinate.cpp.o] Error 1
make[3]: *** Waiting for unfinished jobs....
make[3]: Leaving directory '/var/lib/pgsql/git/pgrpms/rpm/redhat/master/sfcgal/master/SFCGAL-1.3.6'
make[2]: *** [CMakeFiles/Makefile2:95: src/CMakeFiles/SFCGAL.dir/all] Error 2

`

[lrineau]

The code of SFCGAL uses several internal details of CGAL, and those details have changed. SFCGAL is not ready for CGAL-4.14, unfortunately.

[lrineau]

The first step to make the code compile is to use the flag -DCGAL_DO_NOT_USE_BOOST_MP. But that is not sufficient.

@sloriot There is a compilation error in files you added recently:

In file included from /home/lrineau/Git/SFCGAL/src/../CGAL_patches/CGAL/intersection_of_Polyhedra_3.h:46,
                 from /home/lrineau/Git/SFCGAL/src/algorithm/Intersection3D.cpp:24:
/home/lrineau/Git/SFCGAL/src/../CGAL_patches/CGAL/internal/corefinement/intersection_triangle_segment_3.h:30:10: fatal error: CGAL/internal/Intersections_3/Triangle_3_Segment_3_intersection.h: No such file or directory
 #include <CGAL/internal/Intersections_3/Triangle_3_Segment_3_intersection.h>
          ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

[devrimgunduz]

Hi,

This is breaking upgrages to Fedora 30, which is already released. Also, RHEL 8 is out very soon, and this will break more users.

I mean, we need CGAL 4.14 support soon :(

(FWIW, I am packaging RPMs for the PostgreSQL YUM/ZYPP repos)
Regards, Devrim

[fxcoudert]

This is breaking build of SFCGAL in Homebrew as well, now that we have shipped GCC 9: Homebrew/homebrew-core#38890

[fxcoudert]

Given no activity, and it's not the first time we've been blocked by SFCGAL from updating other software (#172) I will propose to Homebrew to remove it. It will, sadly, impact PostGIS functionality… but we currently have no other option.

It's a bit sad that CGAL and its C++ wrapper cannot work together so they're kept in sync.

[sloriot]

Please try: #189

[sloriot]

Together with -DCGAL_DO_NOT_USE_BOOST_MP or set CGAL_WITH_GMPXX:BOOL=ON when configuring SFCGAL.

[lrineau]

@sloriot commented on May 19, 2019, 10:48 AM GMT+2:

Together with -DCGAL_DO_NOT_USE_BOOST_MP or set CGAL_WITH_GMPXX:BOOL=ON when configuring SFCGAL.

I think SFCGAL headers could hardcode:

#define CGAL_DO_NOT_USE_BOOST_MP 1
#define CGAL_DO_NOT_USE_GMPXX 1

The second macro, CGAL_DO_NOT_USE_GMPXX has currently no effect in CGAL, but it will, once I have enabled GMPXX by default (for CGAL-5.0 probably).

[fxcoudert]

I've tried applying the patch at #189 and building with cmake -DCGAL_DO_NOT_USE_BOOST_MP=TRUE. It still gives the same error:

/tmp/sfcgal-20190521-49736-1lujjuk/SFCGAL-1.3.6/src/Coordinate.cpp:276:32: error: no matching function for call to 'round'
                               SFCGAL::round( v.exact() * _scaleFactor ),
                               ^~~~~~~~~~~~~
/tmp/sfcgal-20190521-49736-1lujjuk/SFCGAL-1.3.6/include/SFCGAL/numeric.h:50:15: note: candidate function not viable: no known conversion from 'typename enable_if<is_compatible_arithmetic_type<long, number<gmp_rational, et_on> >, detail::expression<detail::multiply_immediates, number<gmp_rational, et_on>, long> >::type' (aka 'boost::multiprecision::detail::expression<boost::multiprecision::detail::multiply_immediates, boost::multiprecision::number<boost::multiprecision::backends::gmp_rational, boost::multiprecision::et_on>, long, void, void>') to 'const double' for 1st argument
inline double round( const double& v )
              ^
/tmp/sfcgal-20190521-49736-1lujjuk/SFCGAL-1.3.6/include/SFCGAL/numeric.h:86:23: note: candidate function not viable: no known conversion from 'typename enable_if<is_compatible_arithmetic_type<long, number<gmp_rational, et_on> >, detail::expression<detail::multiply_immediates, number<gmp_rational, et_on>, long> >::type' (aka 'boost::multiprecision::detail::expression<boost::multiprecision::detail::multiply_immediates, boost::multiprecision::number<boost::multiprecision::backends::gmp_rational, boost::multiprecision::et_on>, long, void, void>') to 'const CGAL::Gmpq' for 1st argument
SFCGAL_API CGAL::Gmpz round( const CGAL::Gmpq& v ) ;
                      ^

[lrineau]

CGAL_DO_NOT_USE_BOOST_MP is not a CMake variable, but the name of a C++ preprocessor macro.

You can add #define CGAL_DO_NOT_USE_BOOST_MP 1 near the top of the .cpp files of SFCGAL, or add -DCGAL_DO_NOT_USE_BOOST_MP to the compilation flags of the compiler.

[fxcoudert]

a C++ preprocessor macro

OK, thanks. I guess that's the lever of tinkering we're not going to be able to accept for distribution. Multiple patches, plus adding compiler flags (which should be defined in SFCGAL's own headers or build process if they need it)… it's simply too much.

[sloriot]

I'll update the PR to include that.

[devrimgunduz]

Any updates? Fedora 30 users are also asking about this for some time.

[sloriot]

I'm waiting for @mhugo to answer my question in #189. Maybe @vmora can help.

[sloriot]

Thanks @vmora, PR update if you want to give it another try!

[mhugo]

@fxcoudert @devrimgunduz I've merged the fix proposed by @sloriot (sorry for the delay, I was away for a few days). I'll post a new point release if you confirm the last SFCGAL master compiles for you.

[devrimgunduz]

Hi @mhugo ,

This patch fixes the build issue. Thanks!

Waiting for the new point release.

Cheers, Devrim

[mhugo]

@devrimgunduz Thanks for your test. A new release is out: https://github.com/Oslandia/SFCGAL/releases/tag/v1.3.7

[devrimgunduz]

Thanks! Rolling out new packages.

Closing this issue.

Cheers, Devrim