feat(solidlsp): add ls_args and ls_extra_args for custom LS launch command

[arikon]

Summary

• ls_extra_args — extra arguments appended to the default language server launch command (e.g. ["--query-driver=/usr/bin/arm-none-eabi-gcc"])
• ls_args — fully replaces default arguments; only the executable path is kept (e.g. ["--log=verbose"])
• Both options work via ls_specific_settings in .serena/project.yml and apply to all language servers using LanguageServerDependencyProviderSinglePath
• Documented new options in ClangdLanguageServer docstring as an example

Example config

ls_specific_settings:
  cpp:
    ls_extra_args:
      - "--query-driver=/usr/bin/arm-none-eabi-gcc"

[arikon]

@opcode81 Could you have a look on this, please?

[arikon]

Rebased on main

[MischaPanch]

@arikon This is less innocent than it might first seem. The additional launch commands may require an invalidation of the symbols cache. Could you explain which problem you ran into that required this? Maybe it would be sufficient to just extend a single LS implementation before extending the config.

Extending the config like proposed here is a decision that we can't take back once released, so we have to properly think about this and we didn't have time to do so yet

[arikon]

@MischaPanch I'm running clangd из VSCode with such config and a bunch of .clangd files through the tree of my monorepo:

{
        "clangd.arguments": [
            "--enable-config",
            "--header-insertion=never",
            "--log=info",
            "--pretty",
            "-j=6",
            "--clang-tidy",
            "--query-driver=/Users/arikon/.ya/tools/v4/10234711056/bin/arm-none-eabi-g++,/Users/arikon/.ya/tools/v4/10234711056/bin/arm-none-eabi-gcc,/Users/arikon/.ya/tools/v4/10492063598/bin/clang,/Users/arikon/.ya/tools/v4/10492063598/bin/clang++,/Users/arikon/.ya/tools/v4/10592126941/riscv32-esp-elf/bin/riscv32-esp-elf-g++,/Users/arikon/.ya/tools/v4/10592126941/riscv32-esp-elf/bin/riscv32-esp-elf-gcc",
            "--all-scopes-completion"
        ]
}

.clangd in the root of monorepo looks like this:

Index:
  Background: Skip
---

If:
  PathMatch: smart_devices/libs/iot_cloud_sdk/.*
CompileFlags:
  CompilationDatabase: /Users/arikon/ws/vscode/iot_cloud_sdk/compile-commands/default/host
  Add:
    # Workaround: https://github.com/llvm/llvm-project/issues/189288
    - -isystem
    - /Users/arikon/.ya/tools/v4/9996985259/lib/clang/20/include

---

If:
  PathMatch:
    - smart_devices/libs/iot_cloud_sdk/platform/beken/.*
    - smart_devices/third_party/beken_freertos_sdk/.*
    - smart_devices/third_party/freertos/kernel/.*
    - smart_devices/third_party/lwip/.*
CompileFlags:
  CompilationDatabase: /Users/arikon/ws/vscode/iot_cloud_sdk/compile-commands/default/bk7238
  Remove:
    - -mthumb-interwork
    - -mabi=*
    - -mno-sched-prolog
  Add:
    - -Wno-unknown-attributes

---

If:
  PathMatch:
    - smart_devices/libs/iot_cloud_sdk/platform/esp32/.*
    - smart_devices/third_party/espressif/esp-idf/5.4
    - smart_devices/third_party/espressif/esp-idf-aux/5.4
CompileFlags:
  CompilationDatabase: /Users/arikon/ws/vscode/iot_cloud_sdk/compile-commands/default/esp32-esp-idf-5-4
  Add:
    - -Wno-unknown-attributes

The project is — multiplatfrom embedded SDK, so I need to tune clangd like this to make its LSP provide me with correct diagnostics.

I want that clangd ran by Serena will work exactly like clangd in VSCode.

So I need to provide minimun this bunch of args to Serena's clangd:

--enable-config
--clang-tidy
--query-driver=/Users/arikon/.ya/tools/v4/10234711056/bin/arm-none-eabi-g++,/Users/arikon/.ya/tools/v4/10234711056/bin/arm-none-eabi-gcc,/Users/arikon/.ya/tools/v4/10492063598/bin/clang,/Users/arikon/.ya/tools/v4/10492063598/bin/clang++,/Users/arikon/.ya/tools/v4/10592126941/riscv32-esp-elf/bin/riscv32-esp-elf-g++,/Users/arikon/.ya/tools/v4/10592126941/riscv32-esp-elf/bin/riscv32-esp-elf-gcc

[MischaPanch]

clangd like this to make its LSP provide me with correct diagnostics.

Serena doesn't even offer any diagnostics tools now, I don't understand what practically changes for you if Serena runs its current clangd command instead of a customized version. Do you not get proper symbol resolution or are there any problems with performance? Why does it matter if the clangd session that Serena uses doesn't have the exact same parametrization as the one you use in VSC?