Re: [Boost-build] building boost on interix

Subject: Re: [Boost-build] building boost on interix
From: Vladimir Prus (ghost_at_[hidden])
Date: 2009-07-22 11:07:20

On Wednesday 22 July 2009 Robert Oeffner wrote:

> Hi,
> I'm trying to build Boost1.39.0 on interix which is a UNIX subsystem running on top of the NT kernel for Windows. MS provides interix as an optional add-on.
> The particular software I'm trying to compile is the Crystallographic Toolbox, CCTBX, and although it compiles it is with several warnings virtually all of the form:
> 
> boost/python/detail/signature.hpp:77: warning: sorry: semantics of inline function static data 'const boost::python::detail::signature_element result [10]' are wrong (you'll wind up with multiple copies)
> 
> A problem with GCC on interix is that binutils do not support weak symbols. From what I've read I believe that the above warning is due to the compilers lack of support for weak symbols. A small test example on 
> http://www.mail-archive.com/[e-mail]/msg46130.html 
> illustrates the problem. Indeed when I compile this piece of code on interix I get:
> 
> ~/tmp/weakboost>
> ~/tmp/weakboost> g++ -I/E/Users/oeffner/Interix/Boost/boost_1_39_0 weak1.cpp
> In file included from weak1.cpp:2:
> /E/Users/oeffner/Interix/Boost/boost_1_39_0/boost/crc.hpp: In static member function 'static void boost::detail::crc_table_t<Bits, TruncPoly, Reflect>::init_table() [with unsigned int Bits = 32u, typename boost::uint_t<Bits>::fast TruncPoly = 79764919u, bool Reflect = true]':
> /E/Users/oeffner/Interix/Boost/boost_1_39_0/boost/crc.hpp:838:   instantiated from 'boost::crc_optimal<Bits, TruncPoly, InitRem, FinalXor, ReflectIn, ReflectRem>::crc_optimal(typename boost::detail::mask_uint_t<Bits>::fast) [with unsigned int Bits = 32u, typename boost::uint_t<Bits>::fast TruncPoly = 79764919u, typename boost::uint_t<Bits>::fast InitRem = -1u, typename boost::uint_t<Bits>::fast FinalXor = -1u, bool ReflectIn = true, bool ReflectRem = true]'
> weak1.cpp:6:   instantiated from here
> /E/Users/oeffner/Interix/Boost/boost_1_39_0/boost/crc.hpp:492: warning: sorry: semantics of inline function static data 'bool did_init' are wrong (you'll wind up with multiple copies)
> /E/Users/oeffner/Interix/Boost/boost_1_39_0/boost/crc.hpp:492: warning:   you can work around this by removing the initializer
> /tmp/ccCxFr4u.o(.text+0xf0):weak1.cpp: undefined reference to `boost::detail::crc_table_t<32u, 79764919u, true>::table_'
> collect2: ld returned 1 exit status
> ~/tmp/weakboost>
> 
> The compiled binaries from the CCTBX also have problems. It is at present not possible to load boost-python modules from CCTBX in python on interix as it produces a core dump. I think that the lack of support for weak symbols used by the boost library templates  is what causes this.
> However, Boost compiles well on platforms without weak symbol support, such as win32 (the cited example compiles fine with VC2005 as does the CCTBX). I would like to know what flags I can tweak a compilation of Boost with on interix to specify that I do not want GCC attempting to use weak symbols. If anyone has got any suggestions that would be great.

Robert,

unfortunately, I cannot be very helpful here. It does not seem like your problem is that
Boost requires weak symbols support in general -- it is not. Rather, it seems like your
gcc/ld are broken. Technically, gcc has a -fno-weak option, that you can add by

        cxxflags=-fno-weak

option to Boost.Build. However, I never used that personally, and it might break more
than it fix. I suggest you either contract Interix support, or use other gcc,
like mingw/cygwin, or use native windows compiler.

- Volodya