Could not compile libcxx: malloc/malloc.h not found

[aur3l14no]

SymCC is compiled successfully. But building libcxx got me these errors (during configuration).

export SYMCC_REGULAR_LIBCXX=yes
export SYMCC_NO_SYMBOLIC_INPUT=yes

cmake -G Ninja /workspace/llvm-project/llvm \
      -DLLVM_ENABLE_RUNTIMES="libcxx;libcxxabi" \
      -DLLVM_TARGETS_TO_BUILD="X86" \
      -DLLVM_DISTRIBUTION_COMPONENTS="cxx;cxxabi;cxx-headers" \
      -DCMAKE_BUILD_TYPE=Release \
      -DCMAKE_INSTALL_PREFIX=/home/vscode/libcxx_symcc \
      -DCMAKE_C_COMPILER=/workspace/symcc/build/symcc \
      -DCMAKE_CXX_COMPILER=/workspace/symcc/build/sym++

/workspace/symcc/build/symcc   -O3 -DNDEBUG -o CMakeFiles/cmTC_a2a0a.dir/CheckIncludeFile.c.o   -c CheckIncludeFile.c
CheckIncludeFile.c:1:10: fatal error: 'malloc/malloc.h' file not found
#include <malloc/malloc.h>
         ^~~~~~~~~~~~~~~~~
1 error generated.
ninja: build stopped: subcommand failed.

...
fatal error: 'mach/mach.h' file not found
fatal error: 'CrashReporterClient.h' file not found
...

Kindly, any advice on how to fix this error would be appreciated.

[aurelf]

I'm not sure how much the LibCXX instructions are still up to date. If you followed the instructions in docs/C++.txt you used the current git version of the LLVM project? There may be breaking changes. Can you try compiling against LLVM-16 or 15 ? We should support those versions.

[aurelf]

An alternative is to rely on the docker files libcxx seem to work there (although I see there are issues with the symcc_fuzzing_helper but this is another problem)

[aur3l14no]

I actually was using LLVM-16 (16.0.6), which is built and installed via

cmake -DCMAKE_BUILD_TYPE=Release \
      -DLLVM_ENABLE_PROJECTS="clang" \
      -DLLVM_ENABLE_RUNTIMES="libcxx;libcxxabi;libunwind" \
      -DLLVM_TARGETS_TO_BUILD="X86" \
      -G Ninja /workspace/llvm-project/llvm
cmake --build .
cmake --build . --target install

[aur3l14no]

Oh... I realized that I even failed to build cxx with clang. I think it's not an issue with symcc then. Sorry for bothering.

[aurelf]

No problem, I'll close it for now, let us know if you succeed to build the libcxx with SymCC.

[aur3l14no]

Hey @aurelf I'd like to update that I have built libcxx successfully, but with somewhat different configs.

cmake -G Ninja -S /workspace/llvm-project/runtimes \
      -DCMAKE_BUILD_TYPE=Release \
      -DLLVM_TARGETS_TO_BUILD="X86" \
      -DLLVM_ENABLE_RUNTIMES="libcxx;libcxxabi;libunwind" \
      -DCMAKE_INSTALL_PREFIX=/home/vscode/libcxx_symcc \
      -DCMAKE_C_COMPILER=symcc \
      -DCMAKE_CXX_COMPILER=sym++
ninja cxx cxxabi unwind
ninja install-cxx install-cxxabi install-unwind

Please note that the source directory is changed to runtimes. It seems that with llvm as source dir, I can only perform a 2-stage build (build clang first then runtimes), and I can't just stick a compiled clang (symcc) to replace the first stage otherwise there will be errors (maybe related to cmake files but I didn't investigate further).

Anyways, I followed the guide here (https://releases.llvm.org/16.0.0/projects/libcxx/docs/BuildingLibcxx.html#the-default-build) and came up with the above build command. And it worked.

---

Also, I had to tweak this line in sym++, otherwise I would see similar errors as in #104

stdlib_cflags="-isystem ${!libcxx_var}/include/c++/v1 -nostdinc++ -nostdlib++"

Going through all that, I finally managed to symbolize the sample.cpp. However, it failed to generate the alternative input although the constraints did somewhat show the answer...

(assert ...)
(assert ...)
(assert (and (= stdin0 #x72) (= stdin1 #x6f) (= stdin2 #x6f) (= stdin3 #x74)))

# the result is unsat

I noticed in the README that, with QSYM, the output is named "xxx_optimistic". Does it mean the result rely on optimistic (underconstrained) solving to become SAT? I haven't been able to build the QSYM backend so I couldn't test it out.

[aurelf]

Thanks for the investigation ! Reopening. Looks like the port to LLVM 16 was not complete, the CI tests need to be improved in that respect :) Unfortunately, I don't have time to look this more before a while. @sebastianpoeplau if you do, could you have a look ?