Clangd is a language server for C++/C etc. I try to use clangd 12 for C++ code auto-completion on CentOS 7.

My failed attempt to use clang binary release#

Clangd can be easily installed via its binary release. However, when I run clangd --version, I see the following error:

clangd_12.0.0/bin/clangd: /lib64/libc.so.6: version `GLIBC_2.18’ not found (required by clangd_12.0.0/bin/clangd)

The reason is that libc.so on CentOS 7 is too old for clangd to run. We need to use a newer version of glibc (at least glibc-2.18).

I tried to clone glibc repo and build version 2.20 from scratch:

git clone https://mirrors.tuna.tsinghua.edu.cn/git/glibc.git
cd glibc
git checkout glibc-2.20

cd ..
# Note that glibc requires out-source building, i.e., we are forbidden to build glibc inside its source directory.
mkdir glibc-build
../glibc/configure --prefix=$HOME/local
make -j 16
make install

After that, I added $HOME/local/lib to LD_LIBRARY_PATH. Then things went badly wrong: every executable I use segfaults immediately! For example, when I run ls, I got an error from the shell:

zsh: segmentation fault (core dumped) ls -F –color=auto

According to this post, this is due to version mismatch: the glibc version required to run the executable is different from what is provided right now.

I tried the glibc-6.18 release tarball downloaded from here. The same errors happens when I add $HOME/local/lib to LD_LIBRARY_PATH and run clang.

Building clangd from source: success#

I also tried to build llvm project from scratch to use clangd. Here is how to compile llvm from source:

git clone --depth=1 https://github.com/llvm/llvm-project.git

cd llvm-project
mkdir build && cd build
cmake -G "Unix Makefiles" -DLLVM_ENABLE_PROJECTS="clang;clang-tools-extra" -DCMAKE_INSTALL_PREFIX=~/tools/llvm -DCMAKE_BUILD_TYPE=Release ../llvm
make -j 16
make install

Some explanation about the options used:

• -DLLVM_ENABLE_PROJECTS="clang;clang-tools-extra": it specifies which targets we want to build. To use clangd and clang-tidy, clang-tools-extra is a must.
• -DCMAKE_INSTALL_PREFIX=~/tools/llvm: it specifies where we want to install llvm, in this case, we install it under ~/tools/llvm.
• -DCMAKE_BUILD_TYPE=Release: it specifies the build type. Release type will be smaller than Debug.

For more details about Cmake options on building llvm, check here.

Since llvm is a huge project, it may take quite a while to build. FYI, It takes about 37 minutes to build on my server with 16 parallel processes.

After installing llvm, do not forget to add llvm binary to your PATH variable. This time, clangd runs without any error.

How to use clangd#

According to clangd documentation, you can either generate a compile_commands.json file for your project or use compile_flags.txt.

For simple project, using compile_flags.txt is sufficient. Your compile flags are written one per line. A sample compile_flags.txt looks like this:

-Wall
-std=c++11
-L/usr/local/lib
-I/usr/local/include
-lfmt

References#

• glibc segfaults issue:
  • https://unix.stackexchange.com/q/272606/221410
  • https://stackoverflow.com/q/19709932/6064933
• Build clangd: https://github.com/llvm/llvm-project/tree/main/clang-tools-extra/clangd#building-and-testing-clangd