[wiki] Outdated instructions on how to build libcxx instructed with MSAN

[Wentzell]

The wiki instructions on how to build libcxx instructed with MSAN are outdated.
The current article still suggests using LLVM_ENABLE_PROJECTS which is no longer available.

# clone LLVM
git clone --depth=1 https://github.com/llvm/llvm-project
cd llvm-project
mkdir build; cd build
# configure cmake
cmake -GNinja ../llvm \
	-DCMAKE_BUILD_TYPE=Release \
	-DLLVM_ENABLE_PROJECTS="libcxx;libcxxabi" \
	-DCMAKE_C_COMPILER=clang \
	-DCMAKE_CXX_COMPILER=clang++ \
	-DLLVM_USE_SANITIZER=MemoryWithOrigins
# build the libraries
cmake --build . -- cxx cxxabi

Using instead LLVM_ENABLE_RUNTIMES has not been successful for me.
Could you please update the article page for the latest releases?

[yugr]

Just in case, this gist has worked for me a year ago.

[Wentzell]

Thank you @yugr for the suggestion. Unfortunately this does not work for me.
See my setup at https://github.com/TRIQS/triqs/blob/DEV_MSAN/packaging/Dockerfile.msan

The libcxx build fails with a MemorySanitizer positive

110.5 [234/1473] Building IntrinsicEnums.inc...
110.5 FAILED: include/llvm/IR/IntrinsicEnums.inc /tmp/llvm_build/include/llvm/IR/IntrinsicEnums.inc
110.5 cd /tmp/llvm_build && /tmp/llvm_build/bin/llvm-min-tblgen -gen-intrinsic-enums -I /tmp/llvm_src/llvm/include/llvm/IR -I/tmp/llvm_build/include -I/tmp/llvm_src/llvm/include /tmp/llvm_src/llvm/include/llvm/IR/Intrinsics.td --wr
ite-if-changed -o include/llvm/IR/IntrinsicEnums.inc -d include/llvm/IR/IntrinsicEnums.inc.d
110.5 ==1974==WARNING: MemorySanitizer: use-of-uninitialized-value
110.5     #0 0x560ccd4a15a8 in std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char>>::_M_is_local() const /usr/bin/../lib/gcc/x86_64-linux-gnu/12/../../../../include/c++/12/bits/basic_string.h:275:6
110.5     #1 0x560ccd4a15a8 in std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char>>::operator=(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char>>&&) /usr/bin/../lib/gcc/x86_64-l
inux-gnu/12/../../../../include/c++/12/bits/basic_string.h:867:12
110.5     #2 0x560ccd4a15a8 in (anonymous namespace)::CommandLineParser::ParseCommandLineOptions(int, char const* const*, llvm::StringRef, llvm::raw_ostream*, bool) /tmp/llvm_src/llvm/lib/Support/CommandLine.cpp:1520:15
110.5     #3 0x560ccd4a15a8 in llvm::cl::ParseCommandLineOptions(int, char const* const*, llvm::StringRef, llvm::raw_ostream*, char const*, bool) /tmp/llvm_src/llvm/lib/Support/CommandLine.cpp:1469:24
110.5     #4 0x560ccd484cce in main /tmp/llvm_src/llvm/utils/TableGen/TableGen.cpp:82:3
110.5     #5 0x7f86b7615a8f  (/lib/x86_64-linux-gnu/libc.so.6+0x23a8f) (BuildId: bdb8aa3b1b60f9d43e1c70ba98158e05f765efdc)
110.5     #6 0x7f86b7615b48 in __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x23b48) (BuildId: bdb8aa3b1b60f9d43e1c70ba98158e05f765efdc)
110.5     #7 0x560ccd36a844 in _start (/tmp/llvm_build/bin/llvm-min-tblgen+0x42844) (BuildId: edc733bed5cf109b1f4aa4060fc31e5c4b9d77d1)
110.5
110.5   Uninitialized value was created by an allocation of 'ref.tmp12.i' in the stack frame
110.5     #0 0x560ccd4983a1 in (anonymous namespace)::CommandLineParser::ParseCommandLineOptions(int, char const* const*, llvm::StringRef, llvm::raw_ostream*, bool) /tmp/llvm_src/llvm/lib/Support/CommandLine.cpp:1520:17
110.5     #1 0x560ccd4983a1 in llvm::cl::ParseCommandLineOptions(int, char const* const*, llvm::StringRef, llvm::raw_ostream*, char const*, bool) /tmp/llvm_src/llvm/lib/Support/CommandLine.cpp:1469:24
110.5
110.5 SUMMARY: MemorySanitizer: use-of-uninitialized-value /usr/bin/../lib/gcc/x86_64-linux-gnu/12/../../../../include/c++/12/bits/basic_string.h:275:6 in std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char
>>::_M_is_local() const

[yugr]

So what happens here is that build instructions which I provided perform a bootstrap compile - i.e. they first build clang and then use it to build the libraries. Compiler (and its build tools, llvm-tblgen in this case) are memory-sanitized but linked against unsanitized host STL library which results in spurious errors from Msan when compiler/tools are run.

It seems this is a known problem: https://stackoverflow.com/a/73827100/2170527 . Following SO instructions I was able to successfully build libcxx/libcxxabi (needed to install clang-15 first):

$ CC=clang-15 CXX=clang++-15 cmake -G Ninja -S runtimes -B msan_out -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_RUNTIMES="libcxx;libcxxabi" -DLLVM_USE_SANITIZER=MemoryWithOrigins
$ ninja -C msan_out

[vitalybuka]

There is buildbot https://lab.llvm.org/buildbot/#/builders/sanitizer-x86_64-linux-bootstrap-msan

stage2 is libcxx and full clang instrumented and tested with msan

This is how you ca reproduce the bot on your workstation/containter: https://github.com/google/sanitizers/wiki/SanitizerBotReproduceBuild

Or you can try to extract cmake/ninja commands from the bot output https://lab.llvm.org/buildbot/#/builders/74/builds/22184/steps/2/logs/stdio

[yugr]

Or you can try to extract cmake/ninja commands from the bot output https://lab.llvm.org/buildbot/#/builders/74/builds/22184/steps/2/logs/stdio

Looks like it matches the SO answer I linked:

cmake -DLLVM_APPEND_VC_REV=OFF -GNinja -DCMAKE_BUILD_TYPE=Release -DLLVM_CCACHE_BUILD=ON -DLLVM_USE_LINKER=lld -DLLVM_ENABLE_ASSERTIONS=ON -DLLVM_ENABLE_PER_TARGET_RUNTIME_DIR=OFF -DCMAKE_C_COMPILER=/b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm_build0/bin/clang -DCMAKE_CXX_COMPILER=/b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm_build0/bin/clang++ '-DLLVM_ENABLE_RUNTIMES=libcxx;libcxxabi' -DLLVM_USE_SANITIZER=Memory '-DCMAKE_C_FLAGS=-fsanitize=memory  ' '-DCMAKE_CXX_FLAGS=-fsanitize=memory  ' /b/sanitizer-x86_64-linux-bootstrap-msan/build/llvm-project/llvm/../runtimes

[Wentzell]

So what happens here is that build instructions which I provided perform a bootstrap compile - i.e. they first build clang and then use it to build the libraries. Compiler (and its build tools, llvm-tblgen in this case) are memory-sanitized but linked against unsanitized host STL library which results in spurious errors from Msan when compiler/tools are run.

It seems this is a known problem: https://stackoverflow.com/a/73827100/2170527 . Following SO instructions I was able to successfully build libcxx/libcxxabi (needed to install clang-15 first):

$ CC=clang-15 CXX=clang++-15 cmake -G Ninja -S runtimes -B msan_out -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_RUNTIMES="libcxx;libcxxabi" -DLLVM_USE_SANITIZER=MemoryWithOrigins
$ ninja -C msan_out

Thank you for pointing out the SO post. Indeed I was only missing the -S runtimes to avoid bootstrapping the compiler.
It would be great if the wiki could be updated accordingly.

[yugr]

It would be great if the wiki could be updated accordingly.

@kcc @ramosian-glider @eugenis FYI