Make event-log can be turned off

[lorniu]

I found eglot sometimes slower than lsp-mode when I edit Java or C# file.
The completions always delay about 1 seconds or more after every type, but in lsp-mode it's not.

I'm puzzled by this, and at last I found, it's the event-log which is turned on by default, that cause the issue.

There is no convenient way to turn off the event-log, so I have to advice the function, like this:

(defvar jsonrpc-log-event-p nil)

(defun jsonrpc--log-event-advice (f &rest args)
  (if jsonrpc-log-event-p (apply f args)))

(advice-add #'jsonrpc--log-event :around #'jsonrpc--log-event-advice)

And then the performance improvement is huge, completions popup almost no delay after type.

I don't think it's nessasary to alway keep event-log on, but is there any way to turn it off conveniently?

Or is it necessary to add options to give users to choose whether to turn on the event-log?

[joaotavora]

See if this helps:

eglot-events-buffer-size is a variable defined in `eglot.el'.

Its value is 2000000

Control the size of the Eglot events buffer.
If a number, don't let the buffer grow larger than that many
characters.  If 0, don't use an events buffer at all.  If nil,
let the buffer grow forever.

  You can customize this variable.

[back]

[lorniu]

Yes, it works. Thanks, I should read the code carefully.

[skangas]

Is it worth the performance hit to keep it on by default? The alternative would be to add instructions to the bug reporting page for how to turn it on before reporting a bug.

[Eason0210]

Hi @joaotavora ,

To improve performance, I tried to set eglot-events-buffer-size to 0, but the events buffer will still be created when I run M-x eglotwith clangd. Add got the content in EGLOT events as below, and it will grow up when I input characters in my test.cpp buffer:

[stderr] I[09:11:38.996] Apple clangd version 13.1.6 (clang-1316.0.21.2.3)
[stderr] I[09:11:38.997] Features: mac+xpc
[stderr] I[09:11:38.997] PID: 17457
[stderr] I[09:11:38.997] Working directory: /Users/eason/Code/Cpp
[stderr] I[09:11:38.997] argv[0]: /Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clangd
[stderr] I[09:11:38.997] Starting LSP over stdin/stdout
[stderr] I[09:11:38.997] <-- initialize(1)
[stderr] I[09:11:39.002] --> reply:initialize(1) 4 ms
[stderr] I[09:11:39.003] <-- initialized
[stderr] I[09:11:39.004] <-- textDocument/didOpen
[stderr] I[09:11:39.005] Failed to find compilation database for /Users/eason/code/Cpp/test2.cpp
[stderr] I[09:11:39.005] ASTWorker building file /Users/eason/code/Cpp/test2.cpp version 0 with command clangd fallback
[stderr] [/Users/eason/code/Cpp]
[stderr] /Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang /Users/eason/code/Cpp/test2.cpp -resource-dir=/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/lib/clang/13.1.6
[stderr] I[09:11:39.016] <-- workspace/didChangeConfiguration
[stderr] I[09:11:39.538] <-- textDocument/signatureHelp(2)
[stderr] I[09:11:39.538] <-- textDocument/hover(3)
[stderr] I[09:11:39.538] <-- textDocument/documentHighlight(4)
[stderr] I[09:11:39.570] --> reply:textDocument/signatureHelp(2) 31 ms
[stderr] I[09:11:39.581] --> textDocument/publishDiagnostics
[stderr] I[09:11:39.581] --> reply:textDocument/hover(3) 43 ms
[stderr] I[09:11:39.581] --> reply:textDocument/documentHighlight(4) 43 ms

But when I set eglot-events-buffer-size to 1, the EGLOT events buffer will be created with empty content, and will not grow up when in input characters in my test.cpp buffer

I also tested with python with pyright, set to 0 will not create EGLOT events buffer. And I heard that other friends also tried on css、java，Lua, aslo no EGLOT events buffer created with eglot-events-buffer-size set to 1 .

So, looks like it only happened onclangdrecently.

I am not sure if it is a bug or it's designed.

[joaotavora]

I doubt clang's stderr output that is creating any type of "performance problem", but if you want just reduce clangd s error reporting with something like clangd --log=error. I use clangd as well and the slowdowns I notice, I think, are not related to logging at all, just to the inherent difficulty clangd has when parsing large C++ projects.

[Eason0210]

I doubt clang's stderr output that is creating any type of "performance problem"

I also didn't experience performance issues with the stderr log of clangd.

if you want just reduce clangd s error reporting with something like clangd --log=error

Thanks for mention this option. I will try it.

[lorniu]

One disadvantage of using eglot-events-buffer-size to do this is that, it can't turn on/off the event log on the fly, I have to restart eglot completely to make the change take effect. Maybe using some dynamically-bind variable to control the output of jsonrpc-log is a better option.

Advice on the method can make it work ok, but I don't think I am the only man need this, so, maybe a lighter way is needed.

(defvar eglot-log-event-p nil)

(defun jsonrpc--log-event$toggle-event-log (f &rest args)
  (when (and eglot-log-event-p
             (ignore-errors
               (eq (type-of (car args)) 'eglot-lsp-server)))
    (apply f args)))

(advice-add #'jsonrpc--log-event :around #'jsonrpc--log-event$toggle-event-log)

;; Toggle:

(defun im/toggle-eglot-event-log ()
  (interactive)
  (setq eglot-log-event-p (not eglot-log-event-p))
  (message "EGLOG event log is current: %s"
           (if eglot-log-event-p "ON" "OFF")))

Thanks for your amazing work.

[joaotavora]

I just noticed this here after a long time, sorry about that. It's true you need to restart the connection for the variable to take effect, but it's also true that that is as easy as M-x eglot-reconnect.

[casouri]

I would explain that in the doctoring. I just set the variable and was surprised that it has no effect.

[joaotavora]

Yuan Fu ***@***.***> writes:

I would explain that in the doctoring. I just set the variable and was surprised that it has no effect.

Sounds reasonable. I'll make the change unless you beat me to it with a PR. You've got a copyright assignment, don't you? João

[casouri]

I do ;-) I can do it (in a couple of days).