如何評價微軟的Language Server Protocol?

01-12

Protocol Implementations · Microsoft/language-server-protocol Wiki · GitHub
微軟和Codenvy一起推出了這個標準，意在讓editor和IDE以後能夠使用一個通用的組件來做代碼分析
那麼這個protocol有什麼過人之處和不足？今後的發展如何，可否完全取代單獨的編輯器插件（比如YCM）么？

@空明流轉不是早就說過了嗎，M$ 出的任何東西都不應該流行，因為我們要反抗邪惡公司統治世界的企圖。

LSP搞的好絕對是能革IDE命的存在.

Language Server Protocol主要就是解決市面上那麼多編輯器(m)和那麼多語言(n)共存的問題,以前寫編輯器語言插件要寫 m x n 個, LSP出現後只要寫 m + n 個就可以了.

實際使用也是非常爽,只要編輯器裝完 Language Client Plugin, 任何一個新語言只要配個啟動Lanuage Server的命令就有各種補全. 異常神奇.

LSP本身是非常良好的設計,它脫胎於OmniSharp和TypeScript Server,經過實戰檢驗,可以覆蓋一般語言里補全/查錯/找定義/查看類型/的種種功能. 通信用了JSONRPC,可以用TCP或者stdin/stdout傳輸. 把編輯器行為和語言特性用一種通用的方式解耦,這一點可以秒殺市面上所有編輯器和IDE(YCM不能比).

-------------------------

但是現實是骨感的... 微軟邪惡帝國的本質讓他在開源界永世不得翻身(誤

LSP對文檔的抽象是個二維的字元矩陣, 一些古老的編輯器(Vim/Emacs)對算行列的支持不是很好. 所以編輯器插件實現的時候就很麻煩,或者作者會直接省事把整個文件全部重傳給伺服器. 效率不好.

但是更嚴重的問題語言伺服器沒有寫的好的... 這裡Microsoft/language-server-protocol 列出的實現, 要不是功能不全,要不是Bug多, 一般實現代碼也醜陋無比.

這個問題主要原因還是沒有語言團隊的官方支持.(也就是沒語言團隊理會巨硬的proposal). 民間開發者對編譯器/語言/程序的理解實在太弱,就算有LSP這樣的設計思路給到,也沒有辦法做出一個良好的實現.

相反,如果有語言的Core Team支持的話,Language Server就會很神, 目前只有Rust的LSP是Mozilla官方的人寫的, 效果也不錯.

總的來說, LSP目前是有人用的, 但是用戶太少,實現太爛. 光有良好設計,沒有強大社區召喚力push語言實現者實現還是不行的....(估計也就FB/Google能這麼折騰

為什麼都這麼悲觀……

Clang 團隊開始開發支持 LSP 的 ClangD Clang Developers - ClangD, llvm-mirror/clang-tools-extra, rL294291. 根據討論，Clang 團隊似乎想從 libclang 走向 LSP (libclang 就是 YCM 之類用的後端)；

YCM 作者 Val Markovic 對 ClangD 和 LSP 的評價 ([cfe-dev] ClangD)：

&>From the YCM side: I think this is a *fantastic* idea! We were thinking
about pulling out the C++ completion out of ycmd and into a separate server
that ycmd would talk to (much like it works for several other languages
already) but hearing that clang devs are taking this on saves us a bunch of
work!

And it"s great to hear it will follow the common completion API MS is
pushing; we"ll be supporting that API on the back-end of ycmd some time in
the future (and clangd might accelerate that). We"re looking into possibly
supporting it on the front-end as well.

之後也有 Qt 團隊的人表示很感興趣，不過討論了太多細節，太長我就不貼了。

Rust 的 Language Server （RLS）也是緊跟著 LSP 的標準：Move to LSP 3.0 · Issue #178 · rust-lang-nursery/rls

這貨也要配合著這個看

Anders Hejlsberg on Modern Compiler Construction

怎麼寫service部分比較好，高效（不能每次都重建整個樹）。。

===========================================

ps: 最新的的scala編譯器dotty

RMS都說支持你支持不支持？

Neovim 的 issue 列表上很早就出現對於支持 LSP 的討論了，目測已在計劃之內：Language Server Implementation (LSP) · Issue #5522 · neovim/neovim

超級吊！！！

因為 lsp 的存在，所以 VSCode 只是一個遵循了 lsp 的 client 而已。既然只是個 client，以後 VSCode 的牛逼的功能（接近於 IDE 的靜態分析等）幾乎都能在 Emacs / Vim 上實現了，編輯器界終於達到了前所未有的大和諧，喜大普奔！

emacs-lsp/lsp-mode

使用cquery：C++ language server

需要更多人幫忙C++ language server ... Polish before publishing · Issue #30 · jacobdufault/cquery

Higher priority items I"d like to get addressed before publishing to the marketplace

make sure we don"t load cached file if real file is not present
Very make_unique contributes constructor calls
extern const char not included in outline? see ash_switches.h
show warning for compile_commands.json, autogen if possible; maybe git rev-parse origin/master ; or git rev-parse HEAD
Verify when loading project we add to end of queue so existing open files are indexed first
we don"t need to reindex file unless it has timestamp change, ever, try to add metrics to count how many files get updated after sync
reindex file is command line arguments change
max number of code lens results otherwise vscode will hang if the file has 20k+ code lens
Mention how to disable code lens in README
Drop clang_args in favor of pref or move to .cquery or similar
clang arg pref should be platform specific
Only status update every few hundred milliseconds
Reduce output
index perf: don"t run every DoIdMap; stop after 500 or 1000 iterations
Move completion entirely off querydb thread so it never touches querydb
progress indicator needs to account for index threads which are busy (ie, +1 to a job when they remove the item from the queue until it is done being processed)

Misc issues that need to be dealt with at some point

Make warning "launch.workingDirectory" say "cquery.launch.workingDirectory"
std::sting -&> std::string typo fixit does replaces trailing text content/space, it should insert the new string instead of replacing, likely a bug in FixIt generation
Fix show references for function declaration parameters
Code lens causing screen to scroll - Microsoft/vscode#12294
Show warning for old pref names
No reload notification when adding new pref that changes from default value (include.blacklist)
Find ref on lambda parameter
gv on variable should show variables of the same type
Freshen should drop existing index
labeled stack traces on linux
Insert include by typing a symbol (ie, global symbol search)
see if we can set code lens command title to show custom string
Prefer smallest path found for include when scanning
Clickable controls in tooltips - Microsoft/vscode#29076
See if we can integrate commit characters for code completion
Auto-implement doesn"t strip override
Auto-implement select implemented content for fast ctrl-x ctrl-v
Call tree in context menu
Call tree - use icon for base/derived calls
Completion list private with - or similar
Don"t show errors for ie, *.mojom files
Add task to reload compile_commands.json
Call tree should store index lines and look it up when the user requests goto definition
Add task to drop existing index and do a fresh import
Reparse top document - might be possible by hooking directly into extension API and sending a notification $cquery/onView
Call hierarchy; show class name and line, skip column MyMethod (Foobar:3). If no class name show file MyMethod (http://foo.cc:3)
Call hierarchy: sort by file and then by location in that file
Call tree fails on pure-virtual functions
Investigate jemalloc
Do not report global completions in certain situations, ie, o not report global completions in certain situations, ie, Example: type this: class Foo : public Bar {public:
Idea: only do live-indexing after a save, but use completion thread to do it.