Clojurians Log v2

1) the client has a weird behavior where, for the same file, I'm getting multiple didChangeWatchedFiles, one per type, which seems weird. This leads to unnecessary analysis but it's not the most important problem. Is git doing several operations per file when changing branch ?

2) The server receives the didChangeWatchedFiles notification sequentially, and treat each one separately. Now If I have ns a & b that are both required in c, when both a and b change we re-analyze c twice. In my project I have a few namespaces that are really used everywhere. In my testing the same namespace was re-analyzed 62 times when my branch only changed 59 files (125 didChangeWatchedFiles notifications. If I simply batch and de-duplicate the analysis of referenced files the problem goes away.

Here is the code I used in file_management.clj (sorry about the ugly inefficient code, just wanted something that works first try):

(def queue (atom []))

(defn batching-worker
  []
  (future
    (loop []
      (Thread/sleep 1000)
      (when (seq @queue)
        (let [batch @queue
              n-items (count batch)]
          (swap! queue #(vec (drop n-items %)))
          (doseq [[db* v] (group-by :db* batch)]
            (let [uris (into #{} (mapcat :uris v))]
              (analyze-reference-uris! uris db*)
              (run! #(deliver (:response-promise %) ::done) v)))))
      (recur))))

(batching-worker)

(defn analyze-reference-uris!*
  [uris db*]
  (let [p (promise)]
    (swap! queue #(conj % {:uris uris :db* db* :response-promise p}))
    @p))

(ping @lasse.olavi.maatta you can check if this also solves the problem for you)

I have some utility namespaces, which are referenced in over 600 other namespaces in my project, which might explain why this issue haunts me 😅

you can exclude those namespace from re-analysis with lsp-file-watch-ignored-files as a partial solution, at least when rebasing things

yeah, we had multiple improvements in clojure-lsp related to that, but your fix seems to be a good improvement, feel free to open a pr @rolthiolliere so I can take a closer look

Great if a batching approach seems sound I'll prepare something. The "patch" I wrote here has a few issues.

it's way important to analyze files changed outside editor slowly, but consistenly , than faster and wrong or affecting performance