History and Conflict

Language Server is a cool idea, that (as far as I know) grew out of VS Code, and provides a way to give a text editor "advanced, IDE-like" features without having to implement those features per-language in the text editor. Instead, editor developers implement generic integration for the protocol, and language developers produce a single process that can process source code and supports the editor features. Sounds great!

There are challenges:

• The LSP protocol uses JSON for encoding data between the server and emacs. This is a reasonable choice, and makes development easy, but it's not a perfect fit. Primarily, in emacs, until version 27, which hasn't reached an official release yet (but is quite usable on Linux), parsed JSON in lisp rather than C, which made everything rather slow, but emacs 27 helps a lot. Second,

• There is some overhead for the lsp, particularly as you attempt to scale very large numbers of files or very large files. Emacs is really good with large files, as an artifact of having had to support editing larger files on computers 20 or 30 years ago, and lsp feels sluggish in these situations.

  I think some of this overhead is probably depends on the implementation of the language server's themselves, and some percentage is probably protocol itself: http's statelessness is great in big distributed web services, and for the local use cases where responsiveness matters, it's less good. Similarly JSON's transparency and ubiquity is great, but for streaming streaming-type use cases, perhaps less so.

I've not poked around the protocol, or any implementations, extensively, so I'm very prepared to be wrong about these points.

There's also some tension about moving features, like support for adding editor support for a language, out of lisp and into an external process. While I'm broadly sympathetic, and think there are a bunch of high-quality language integrations in emacs lisp, and for basic things, the emacs lisp ones might be faster. Having said that, there's a lot of interface inconsistency between different support for languages in emacs, and LSP support feels spiffy my comparisons. Also at least for python (jedi,) go (gocode), and C/C++ (irony,) lisp (slime,) the non-lsp services were already using external processes for some portion of their functionality.

It's also the case that if people spend their emacs-lisp-writing time doing things other than writing support for various development environments it's all for the better anyway.

Suggestions / Notes

• things will break or get wedged, call lsp-restart-workspace to restart the server (I think.) It helps.
• use lsp-file-watch-ignored to avoid having the language servers process emacs process files from vendored code, intermediate/generated code, or similar, in cases where you might have a large number of files that aren't actually material to your development work
• I believe some (many? all?) servers are sourced from the PATH (e.g. clang/llvm, gopls), which means you are responsible for updating them and their dependencies.
• I've found it useful to update lsp-mode itself more often than I typically update other emacs packages. I also try and keep the language servers as up to date as possible in coordination with lsp-mode updates.
• If Emacs 27 can work on your system, it's noticeably faster. Do that.
• If you have an existing setup for a language environment, the LSP features end up layering over existing functionality, and that can have unexpected results.

My Configuration

None of this is particularly exciting, but if you use use-package, then the following might be an interesting starting point. I stole a lot of this from other places, so shrug.

I must say that I don't really use dap or treemacs, because I tend to keep things pretty barebones.

(use-package helm-lsp
  :ensure t
  :after (lsp-mode)
  :commands (helm-lsp-workspace-symbol)
  :init (define-key lsp-mode-map [remap xref-find-apropos] #'helm-lsp-workspace-symbol))

(use-package lsp-mode
  :diminish (lsp-mode . "lsp")
  :bind (:map lsp-mode-map
    ("C-c C-d" . lsp-describe-thing-at-point))
  :hook ((python-mode . #'lsp-deferred)
    (js-mode . #'lsp-deferred)
    (go-mode-hook . #'lsp-deferred))
  :init
  (setq lsp-auto-guess-root t       ; Detect project root
   lsp-log-io nil
   lsp-enable-indentation t
   lsp-enable-imenu t
   lsp-keymap-prefix "C-l"
   lsp-file-watch-threshold 500
   lsp-prefer-flymake nil)      ; Use lsp-ui and flycheck

  (defun lsp-on-save-operation ()
    (when (or (boundp 'lsp-mode)
         (bound-p 'lsp-deferred))
      (lsp-organize-imports)
      (lsp-format-buffer))))

(use-package lsp-clients
  :ensure nil
  :after (lsp-mode)
  :init (setq lsp-clients-python-library-directories '("/usr/local/" "/usr/")))

(use-package lsp-ui
  :ensure t
  :after (lsp-mode)
  :commands lsp-ui-doc-hide
  :bind (:map lsp-ui-mode-map
         ([remap xref-find-definitions] . lsp-ui-peek-find-definitions)
         ([remap xref-find-references] . lsp-ui-peek-find-references)
         ("C-c u" . lsp-ui-imenu))
  :init (setq lsp-ui-doc-enable t
         lsp-ui-doc-use-webkit nil
         lsp-ui-doc-header nil
         lsp-ui-doc-delay 0.2
         lsp-ui-doc-include-signature t
         lsp-ui-doc-alignment 'at-point
         lsp-ui-doc-use-childframe nil
         lsp-ui-doc-border (face-foreground 'default)
         lsp-ui-peek-enable t
         lsp-ui-peek-show-directory t
         lsp-ui-sideline-update-mode 'line
         lsp-ui-sideline-enable t
         lsp-ui-sideline-show-code-actions t
         lsp-ui-sideline-show-hover nil
         lsp-ui-sideline-ignore-duplicate t)
  :config
  (add-to-list 'lsp-ui-doc-frame-parameters '(right-fringe . 8))

  ;; `C-g'to close doc
  (advice-add #'keyboard-quit :before #'lsp-ui-doc-hide)

  ;; Reset `lsp-ui-doc-background' after loading theme
  (add-hook 'after-load-theme-hook
       (lambda ()
         (setq lsp-ui-doc-border (face-foreground 'default))
         (set-face-background 'lsp-ui-doc-background
                              (face-background 'tooltip))))

  ;; WORKAROUND Hide mode-line of the lsp-ui-imenu buffer
  ;; @see https://github.com/emacs-lsp/lsp-ui/issues/243
  (defadvice lsp-ui-imenu (after hide-lsp-ui-imenu-mode-line activate)
    (setq mode-line-format nil)))

;; Debug
(use-package dap-mode
  :diminish dap-mode
  :ensure t
  :after (lsp-mode)
  :functions dap-hydra/nil
  :bind (:map lsp-mode-map
         ("<f5>" . dap-debug)
         ("M-<f5>" . dap-hydra))
  :hook ((dap-mode . dap-ui-mode)
    (dap-session-created . (lambda (&_rest) (dap-hydra)))
    (dap-terminated . (lambda (&_rest) (dap-hydra/nil)))))

(use-package lsp-treemacs
  :after (lsp-mode treemacs)
  :ensure t
  :commands lsp-treemacs-errors-list
  :bind (:map lsp-mode-map
         ("M-9" . lsp-treemacs-errors-list)))

(use-package treemacs
  :ensure t
  :commands (treemacs)
  :after (lsp-mode))

Org Mode Guilt and a Lisp Function

I have some guilt about having mostly forsaken org-mode, [1] in particular because I was watching Sacha Chua's chat with John Wiegley, and I think both are such nifty hackers, and have done so many things that are pretty darn nifty.

I liked what I heard about johnw's org mode setup so much that I might give it a try again. But in the mean time, I wanted to make my "recompile my tasklist function" to be a bit more clean. The result is follows:

(defun tychoish-todo-compile ()
   (interactive)
   (if (get-buffer "*todo-compile*")
       (progn
          (switch-to-buffer-other-window (get-buffer "*todo-compile*"))
          (recompile))
       (progn
          (compile "make -j -k -C ~/wiki")
          (switch-to-buffer-other-window "*compilation*")
          (rename-buffer "*todo-compile*")))
       (revbufs))

Notables:

• This is the first time I've used progn which is somewhat embarrassing, but it's a great thing to have in the toolkit now. Link: progn
• I hadn't realized until now that there wasn't an else-if form in emacs lisp. Weird, but it makes sense.
• Compilation Mode is pretty much my current favorite thing in emacs.
• revbufs is this amazing thing that reverts buffers if there aren't local modifications, and also reports to you if a buffer has changed outside of emacs and there are local modifications. So basically "does everything you want without destroying anything and then tells you what you need to do manually." Smart. Simple. Perfect.

I might need to "macro-ize" this, as I have a lot of little compile processes for which I'd like to be able to trigger/maintain unique compile buffers. That's a project for another day.

Emacs Thoughts

I'm even thinking about putting together a post about how, although I'm a diehard emacs user, and I've spent a fair bit of time learning how to do really great things with emacs, there are a lot of vim-ish things in my workflow:

• I read email with mutt and I've tried to get into GNUS, and I try it again every now and then, but I always find it so unbelievably gnarly. At least the transition. Same with Notmuch, which I like a lot more (in theory,) but I find the fact that Notmuch and mutt have this fundamental misunderstanding about what constitutes a "read" email, to be tragic.

• I use a crazy ikiwiki + deft + makefile setup for task tracking. As (obliquely) referenced above.

  I might give org another shot, and I've been looking at task warrior, but the sad truth is that what I have works incredibly well for in most cases, and switching is hard.

• I tend jump to a shell window to do version control and other things, even though I'm familiar with magit and dired, my use of these tools is somewhat spotty.

The Perils of Org

The problem I keep running into with org is that I really don't prefer to work in org-mode. [1] Org is great and very flexible, but I don't like that it means that all text-based work is dependent on emacs. My brain is already wired for Markdown and reStructured Text from years of blogging and work projects respectively.

And then there's this organization problem. There are two ways you can organize content in org-mode. The first is to just dump every thing in one org-mode file and use the hierarchical outlining to impose organization to organize everything. The second is to have every project inside of it's own file and use outlining incidentally as the project needs it. Content aggregation happens in the agenda.

The problem with the "large files" approach is that you end up with a small handful of files with thousands of lines and imposing useful organization is difficult (too many levels and things get buried; not enough and inevitably your headings aren't descriptive enough and you get confused. Furthermore, I end up living in clone-indirect-buffer-other-window'd and org-narrow-to-subtree'd buffers, which is operationally the same as having multiple files it just takes longer to set up.

The problem with the other approach, having lots of different files, is that I have a hard time remembering what is in each file, or in logically splitting big projects into multiple files. The agenda does help with this, but the truth is that the kinds of org-headings for organization and tasks are not always the same kinds of headings that make sense for the project itself. I often need more tasks than organizational divides in a project. I tried this approach a couple of times, and ended up with useless mush in my files.

Typically, I can never make the "lots of file approach" really work, and the big files problem lead me to general avoidance of everything. Not good. The key to success here is good aggregation tools.

Hodgepodge

In response, I've made a couple of tweaks to how I'm doing... pretty much everything. That is:

• I've moved most of my open projects into a locally ruining and compiling ikiwiki instance. Both laptops have this setup, and there's a central remote to keep both (all?) machines in sync.
• I'm using ikiwiki tasklist to basically replicate the functions of org-agenda. Basically this crawls the entire wiki looking for lines that begin with certain keywords and generates a "todo" page based on these notes. Really simple, incredibly useful and it solves much of my aggregation needs.
• I still have some stuff in org-mode: notes for the nearly-finished novel, lots of random old (legacy) data, 12 various open tasks, and org-capture. I'm thinking of pointing various org-capture templates at files in the wiki but haven't gotten there yet.
• I've basically taken the "lots of little files," approach to my writing and work. I've not over-leaded the system yet. Each major project gets a page in the root level of the wiki for overview and planing, and then sub-pages for all related project files (if/as needed)
• It turns out that the markdown-mode for emacs has gotten a few improvements since the last time I downloaded the file, including better support for wiki-links that are mostly compatible with ikiwiki. Also from the same developer deft which implements a pretty nifty incremental search for text files in a given directory. So between these tools, ikiwiki, and the ikiwiki-tasklist there's support for the most important things.
• In terms of publishing, beyond ikiwiki for tychoish.com and the personal organization instance, I have a couple of other smaller wikis (also ikiwiki powered,) and I've been playing with Sphinx as publishing for more structured documents and resources (i.e. documentation, novels, and collections,) particularly those that need multiple formats and presentations.

I'm sure there will be more shifts in the future, I'm sure. I think this is a good start. Thoughts?