make[4]: Entering directory `/home/regehr/z/tmp/llvm-r80385/tools/clang/lib/Basic'
llvm[4]: Compiling Builtins.cpp for Release+Checks build
llvm[2]: Linking Release+Checks executable llvm-db (without symbols)
In file included from /usr/include/c++/4.2/debug/debug.h:141,
from /usr/include/c++/4.2/bits/stl_algobase.h:78,
from /usr/include/c++/4.2/algorithm:65,
from /home/regehr/z/tmp/llvm-r80385/include/llvm/ADT/StringRef.h:13,
from /home/regehr/z/tmp/llvm-r80385/include/llvm/ADT/StringMap.h:17,
from /home/regehr/z/tmp/llvm-r80385/tools/clang/lib/Basic/../../include/clang/Basic/IdentifierTable.h:20,
from Builtins.cpp:15:
/usr/include/c++/4.2/debug/formatter.h: In constructor ‘__gnu_debug::_Error_formatter::_Parameter::_Parameter(const __gnu_debug::_Safe_iterator<_Iterator, _Sequence>&, const char*, __gnu_debug::_Error_formatter::_Is_iterator)’:
/usr/include/c++/4.2/debug/formatter.h:218: error: cannot use typeid with -fno-rtti
-fno-rtti wasn't used in the build of clang, it started being used in
the build after a Makefile fix:
I see. Funny no one caught this before.
What should be done about the clang-x86_64-linux-check buildbot, install
that libstdc++ patch?
I don't like changing anything on the buildbot system, part of the
value of keeping all configuration in the buildbot / host is that
installing / replacing a slave is just a matter of installing
buildbot.
Would it be terrible to just build with RTTI for expensive checks builds?