While looking at a qemu issue, I noticed that my poudriere jail for aarch64 has a nxb-bin/usr/lib/clang/11.0.0/lib/freebsd directory, but it's empty, leading to a build failure like: [00:00:40] /usr/local/bin/ld: cannot find /nxb-bin/usr/lib/clang/11.0.0/lib/freebsd/libclang_rt.profile-aarch64.a: No such file or directory Simply copying the usr/lib/clang/11.0.0/lib/freebsd/libclang_rt.profile-aarch64.a to it seems to solve the issue. I don't understand the native-xtools-install target well enough to know why it isn't copied right now, maybe someone can take a look?
lib/Makefile has: # The libraries under libclang_rt can only be built by clang, and only make # sense to build when clang is enabled at all. Furthermore, they can only be # built for certain architectures. .if ${MK_CLANG} != "no" && ${COMPILER_TYPE} == "clang" && \ (${MACHINE_CPUARCH} == "aarch64" || ${MACHINE_CPUARCH} == "amd64" || \ ${MACHINE_CPUARCH} == "arm" || ${MACHINE_CPUARCH} == "i386" || \ ${MACHINE_CPUARCH} == "powerpc") _libclang_rt= libclang_rt .endif while the top-level Makefile.inc1 has: native-xtools: .PHONY ${_+_}cd ${.CURDIR}; ${NXBTMAKE} _cleanobj # Build the bootstrap/host/cross tools that produce native binaries ${_+_}cd ${.CURDIR}; ${NXBTMAKE} kernel-toolchain # Populate includes/libraries sysroot that produce native binaries. # This is split out from 'toolchain' above mostly so that target LLVM # libraries have a proper LLVM_DEFAULT_TARGET_TRIPLE without # polluting the cross-compiler build. The LLVM/GCC libs are skipped # here to avoid the problem but are kept in 'toolchain' so that # needed build tools are built. ${_+_}cd ${.CURDIR}; ${NXBTMAKE} _includes MK_CLANG=no ${_+_}cd ${.CURDIR}; ${NXBTMAKE} _libraries MK_CLANG=no [...] It looks like the MK_CLANG=no was added on purpose, precisely to avoid populating the lib/clang/x.y.z directory. I do not fully understand the reasoning mentioned in the comment about "avoiding the problem". Bryan added this part in base r325001, so maybe he can shed some light.