config CC_GCC_ENABLE_TARGET_OPTSPACE bool prompt "Optimize gcc libs for size" default y help Pass --enable-target-optspace to crossgcc's configure. This will compile crossgcc's libs with -Os. config CC_STATIC_LIBSTDCXX bool prompt "Link libstdc++ statically into the gcc binary" default y depends on CC_GCC_4_4_or_later help Newer gcc versions use the PPL library which is C++ code. Statically linking libstdc++ increases the likeliness that the gcc binary will run on machines other than the one which it was built on, without having to worry about distributing the matching version of libstdc++ along with it. comment "Misc. obscure options." config CC_CXA_ATEXIT bool prompt "Use __cxa_atexit" default y depends on ! BARE_METAL help If you get the missing symbol "__cxa_atexit" when building C++ programs, you might want to try disabling this option. config CC_GCC_DISABLE_PCH bool prompt "Do not build PCH" default n help Say 'y' here to not use Pre-Compiled Headers in the resulting toolchain. at the expense of speed when compiling C++ code. For some configurations (most notably canadian?), PCH are broken, and need to be disabled. Please see: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40974 config CC_GCC_SJLJ_EXCEPTIONS tristate prompt "Use sjlj for exceptions" depends on ! BARE_METAL default m help 'sjlj' is short for setjmp/longjmp. On some architectures, stack unwinding during exception handling works perfectly well without using sjlj, while on some others, use of sjlj is required for proper stack unwinding. Option | sjlj use | Associated ./configure switch ---------+--------------------+-------------------------------- Y | forcibly used | --enable-sjlj-exceptions M | auto | (none, ./configure decides) N | forcibly not used | --disable-sjlj-exceptions It should be safe to say 'M' or 'N'. It can happen that ./configure is wrong in some cases. Known case is for ARM big endian, where you should say 'N'. config CC_GCC_LIBMUDFLAP bool prompt "Compile libmudflap" default n help libmudflap is a pointer-use checking tool, which can detect various mis-usages of pointers in C and (to some extents) C++. You should say 'N' here, as libmduflap generates instrumented code (thus it is a bit bigger and a bit slower) and requires re-compilation and re-link, while it exists better run-time alternatives (eg. DUMA, dmalloc...) that need neither re- compilation nor re-link.