Use Language Servers for Development in the FreeBSD Src Tree

trademarks

FreeBSD is a registered trademark of the FreeBSD Foundation.

1. Introduction

This guide is about setting up a FreeBSD src tree with language servers performing source code indexing. The guide describes the steps for Vim/NeoVim and VSCode. If you use a different text editor you can use this guide as a reference and search the equivalent commands for your preferred editor.

2. Requirements

In order to follow this guide we need to install certain requirements. We need a Language server, ccls or clangd, and optionally a compilation database.

The installation of the Language server can be performed via pkg or via ports. If we chose clangd we need to install llvm.

Using pkg to install ccls:

# pkg install ccls

If we want to use clangd we need to install llvm (The example command uses llvm15 but choose the version you desire):

# pkg install llvm15

To install via ports choose a favorite combination of tools from each category below:

3. Editor settings

3.1. Vim/Neovim

3.1.1. LSP client plugins

The built-in plugin manager is used for both editors in this example. The LSP client plugin used is prabirshrestha/vim-lsp.

To set up the LSP client plugin for Neovim:

# mkdir -p ~/.config/nvim/pack/lsp/start
# git clone https://github.com/prabirshrestha/vim-lsp ~/.config/nvim/pack/lsp/start/vim-lsp

and for Vim:

# mkdir -p ~/.vim/pack/lsp/start
# git clone https://github.com/prabirshrestha/vim-lsp ~/.vim/pack/lsp/start/vim-lsp

To enable the LSP client plugin in the editor, add the following snippet into ~/.config/nvim/init.vim when using Neovim, or ~/.vim/vimrc when using Vim:

For ccls
au User lsp_setup call lsp#register_server({
    \ 'name': 'ccls',
    \ 'cmd': {server_info->['ccls']},
    \ 'allowlist': ['c', 'cpp', 'objc'],
    \ 'initialization_options': {
    \     'cache': {
    \         'hierarchicalPath': v:true
    \     }
    \ }})
For clangd
au User lsp_setup call lsp#register_server({
    \ 'name': 'clangd',
    \ 'cmd': {server_info->['clangd15', '--background-index', '--header-insertion=never']},
    \ 'allowlist': ['c', 'cpp', 'objc'],
    \ 'initialization_options': {},
    \ })

Depending on the version that you installed for clangd you might need to update the server-info to point to the correct binary.

Please refer to https://github.com/prabirshrestha/vim-lsp/blob/master/README.md#registering-servers to learn about setting up key bindings and code completion. The official site of clangd is https://clangd.llvm.org, and the repository link of ccls is https://github.com/MaskRay/ccls/.

Below are the reference settings of keybindings and code completions. Put the following snippet into ~/.config/nvim/init.vim, or ~/.vim/vimrc for Vim users to use it:

function! s:on_lsp_buffer_enabled() abort
    setlocal omnifunc=lsp#complete
    setlocal completeopt-=preview
    setlocal keywordprg=:LspHover

    nmap <buffer> <C-]> <plug>(lsp-definition)
    nmap <buffer> <C-W>] <plug>(lsp-peek-definition)
    nmap <buffer> <C-W><C-]> <plug>(lsp-peek-definition)
    nmap <buffer> gr <plug>(lsp-references)
    nmap <buffer> <C-n> <plug>(lsp-next-reference)
    nmap <buffer> <C-p> <plug>(lsp-previous-reference)
    nmap <buffer> gI <plug>(lsp-implementation)
    nmap <buffer> go <plug>(lsp-document-symbol)
    nmap <buffer> gS <plug>(lsp-workspace-symbol)
    nmap <buffer> ga <plug>(lsp-code-action)
    nmap <buffer> gR <plug>(lsp-rename)
    nmap <buffer> gm <plug>(lsp-signature-help)
endfunction

augroup lsp_install
    au!
    autocmd User lsp_buffer_enabled call s:on_lsp_buffer_enabled()
augroup END

3.2. VSCode

3.2.1. LSP client plugins

LSP client plugins are required to launch the language server daemon. Press Ctrl+Shift+X to show the extension online search panel. Enter llvm-vs-code-extensions.vscode-clangd when running clangd, or ccls-project.ccls when running ccls.

Then, press Ctrl+Shift+P to show the editor commands palette. Enter Preferences: Open Settings (JSON) into the palette and hit Enter to open settings.json. Depending on the language server implementations, put one of the following JSON key/value pairs in settings.json:

For clangd
[
    /* Begin of your existing configurations */
    ...
    /* End of your existing configurations */
    "clangd.arguments": [
        "--background-index",
        "--header-insertion=never"
    ],
    "clangd.path": "clangd12"
]
For ccls
[
    /* Begin of your existing configurations */
    ...
    /* End of your existing configurations */
    "ccls.cache.hierarchicalPath": true
]

4. Compilation database

A Compilation database contains an array of compile command objects. Each object specifies a way of compiling a source file. The compilation database file is usually compile_commands.json. The database is used by language server implementations for indexing purpose.

Please refer to https://clang.llvm.org/docs/JSONCompilationDatabase.html#format for details on the format of the compilation database file.

4.1. Generators

4.1.1. Using scan-build-py

4.1.1.1. Installation

intercept-build tool from scan-build-py is used to generate compilation database.

Install devel/python to get python interpreter first. To get intercept-build from LLVM:

# git clone https://github.com/llvm/llvm-project /path/to/llvm-project

where /path/to/llvm-project/ is your desired path for the repository. Make an alias in the shell configuration file for convenience:

alias intercept-build='/path/to/llvm-project/clang/tools/scan-build-py/bin/intercept-build'

rizsotto/scan-build can be used instead of LLVM’s scan-build-py. The LLVM’s scan-build-py was rizsotto/scan-build merged into the LLVM tree. This implementation can be installed by pip install --user scan-build. The intercept-build script is in ~/.local/bin by default.

4.1.1.2. Usage

In the top-level directory of the FreeBSD src tree, generate the compilation database with intercept-build:

# intercept-build --append make buildworld buildkernel -j`sysctl -n hw.ncpu`

The --append flag tells the intercept-build to read an existing compilation database (if a compilation database exists) and append the results to the database. Entries with duplicated command keys are merged. The generated compilation database by default is saved in the current working directory as compile_commands.json.

4.1.2. Using devel/bear

4.1.2.1. Usage

In the top-level directory of the FreeBSD src tree, to generate compilation database with bear:

# bear --append -- make buildworld buildkernel -j`sysctl -n hw.ncpu`

The --append flag tells bear to read an existing compilation database if it is present, and append the results to the database. Entries with duplicated command keys are merged. The generated compilation database by default is saved in the current working directory as compile_commands.json.

5. Final

Once the compilation database is generated, open any source files in the FreeBSD src tree and LSP server daemon will be launched as well in background. Opening source files in the src tree for the first time takes significantly longer time before the LSP server is able to give a complete result, due to initial background indexing by the LSP server compiling all the listed entries in the compilation database. The language server daemon however does not index the source files not appearing in the compilation database, thus no complete results are shown on source files not being compiled during the make.


Last modified on: December 21, 2023 by rilysh