clang-3.8+ math.h fails to work on Snow Leopard without -U__STRICT_ANSI__

[kenneth.f.cunningham@…]

As Jeremy has pointed out previously, there is a build error in ld64-latest @264.3.102_2+llvm38 on 10.6. due to math errors.

the near-final line of the LibCxxOnOlderSystems instructions

sudo port -v upgrade --enforce-variants ld64 -ld64_97 -ld64_127 -ld64_136 -ld64_236

fails due to some math errors

opt/local/bin/clang-mp-3.8 -Os -Os -arch x86_64 -I/opt/local/libexec/llvm-3.8/include -DLTO_SUPPORT -Isrc/abstraction -Isrc/ld -Isrc/ld/parsers -I/opt/local/include -I/opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_release_tarballs_ports_devel_ld64/ld64-latest/work/dyld-360.18/include  -c -o src/ld/debugline.o src/ld/debugline.c
In file included from src/other/ObjectDump.cpp:35:
In file included from src/ld/parsers/macho_relocatable_file.h:28:
In file included from src/ld/ld.hpp:30:
/opt/local/libexec/llvm-3.8/bin/../include/c++/v1/math.h:1199:91: error: use of undeclared identifier 'llrintf'
inline _LIBCPP_INLINE_VISIBILITY long long llrint(float __lcpp_x) _NOEXCEPT       {return llrintf(__lcpp_x);}
                                                                                          ^
/opt/local/libexec/llvm-3.8/bin/../include/c++/v1/math.h:1200:91: error: use of undeclared identifier 'llrintl'
inline _LIBCPP_INLINE_VISIBILITY long long llrint(long double __lcpp_x) _NOEXCEPT {return llrintl(__lcpp_x);}
                                                                                          ^
/opt/local/libexec/llvm-3.8/bin/../include/c++/v1/math.h:1205:40: error: call to 'llrint' is ambiguous
llrint(_A1 __lcpp_x) _NOEXCEPT {return llrint((double)__lcpp_x);}
                                       ^~~~~~
/opt/local/libexec/llvm-3.8/bin/../include/c++/v1/math.h:1199:44: note: candidate function
inline _LIBCPP_INLINE_VISIBILITY long long llrint(float __lcpp_x) _NOEXCEPT       {return llrintf(__lcpp_x);}
                                           ^
/opt/local/libexec/llvm-3.8/bin/../include/c++/v1/math.h:1200:44: note: candidate function
inline _LIBCPP_INLINE_VISIBILITY long long llrint(long double __lcpp_x) _NOEXCEPT {return llrintl(__lcpp_x);}

-- and more of the same ----

this potential fix was located <​http://clang-developers.42468.n3.nabble.com/problems-building-libcxx-td2353619.html>

and indeed adding the following to the ld64 portfile does enable the ld64 upgrade to proceed without errors, and subsequent ports build normally. Presumably this is 10.6 specific (not tested on any other system at present) and would need to be wrapped in a darwin 10 guard...

    configure.cppflags-append -I${workpath}/dyld-${dyld_version}/include
  +  configure.cppflags-append -U__STRICT_ANSI__

[jeremyhu (Jeremy Huddleston Sequoia)]

Yeah, this is the issue I was mentioning on the mailing list about why I don't recommend using macports-clang-3.8 yet as a default compiler on Snow Leopard. I prefer to get this fixed in the clang-3.8+ ports.

[jeremyhu (Jeremy Huddleston Sequoia)]

3004-Fix-missing-long-long-math-prototypes-when-using-the.patch fixes this for cmath, but v1/math.h is new in 3.8, so we need to fix it as well.

[jeremyhu (Jeremy Huddleston Sequoia)]

To github.com:macports/macports-ports.git

6bcc0e3..b41cdad master -> master

​https://github.com/jeremyhu/libcxx/commit/720feba4874d4ca3131753dc31af127e3c509c36