Should cider work with `--with-poll` emacs?

[jmckitrick]

There's a newish feature in emacs that swaps out select for poll to allow over 1024 file descriptors to be open, IIUC. It's supposedly a 'drop in' replacement. However, my first attempt to run emacs and cider with this option resulted in a hung connection during 'jack in'. Should this feature even work with cider at all? Or is it worth investigating?

[vemv]

Thanks for bringing attention over this!

We could expand our CI matrix and see which combinations succeed/fail.

(There aren't a lot of 'jack in' integration tests, but maybe a more basic test fails, giving us a hint)

[jmckitrick]

I build and run emacs HEAD every week and tinker with settings, so I'm happy to help with this.

[vemv]

I'll try out https://github.com/d12frosted/homebrew-emacs-plus which makes this option easy, apparently

(TIL about that distro)

[jmckitrick]

@vemv I use that one and https://github.com/daviderestivo/homebrew-emacs-head

[vemv]

I ran brew install emacs-head@30 --with-cocoa --with-poll and could repro the issue.

By running M-x toggle-debug-on-error and then cider-connecting I could see

• First:

[nREPL] Establishing direct connection to 127.0.0.1:64844 ...
[nREPL] Direct connection to 127.0.0.1:64844 established

• But later:

  signal(error ("Sync nREPL request timed out (op clone id 1 time-stamp 2023-09-26 16:32:21.439826000) after 10 secs"))
  error("Sync nREPL request timed out %s after %s secs" ("op" "clone" "id" "1" "time-stamp" "2023-09-26 16:32:21.439826000") 10)
  nrepl-send-sync-request(("op" "clone" "id" "1" "time-stamp" "2023-09-26 16:32:21.439826000") #<buffer *cider-uninitialized-repl*> nil nil)
  nrepl-sync-request:clone(#<buffer *cider-uninitialized-repl*>)
  nrepl--init-client-sessions(#<process nrepl-connection>)
  nrepl-start-client-process("127.0.0.1" 64844 nil #f(compiled-function (_) #<bytecode 0xf69663072ad1ccf>) nil)
  cider-nrepl-connect((:host "127.0.0.1" :port 64844 :project-dir "/Users/vemv/orchard/" :repl-init-function nil :session-name nil :repl-type clj))

This would require some investigation.

Perhaps the Emacs 30 changelog explains an adoption guide? i.e. the transition steps.

[jmckitrick]

@vemv I saw exactly the same thing. Yet every article I read on 'poll vs select' claimed it was a drop-in replacement that should be transparent to the user. At least, that was my impression.

I'll take a closer look at the emacs docs, but I'm pretty sure there was no migration guide or something similar.

[vemv]

It might be even an unofficial patch https://github.com/daviderestivo/homebrew-emacs-head/blob/master/patches/0011-Poll.patch - do you know where it comes from?

[jmckitrick]

@vemv Though I wouldn't be surprised if just a handful of simple changes in cider-nrepl solve the entire issue.

[vemv]

Those would be welcome as long that nothing was impacted for the vanilla patch path.

[jmckitrick]

@vemv The patch is for the macos build... but the feature itself is in the underlying emacs. Let me see if I can find it.

[jmckitrick]

But you're right.... this patch is more extensive than I initially thought. Maybe I'll try a vanilla terminal mode emacs and see if that works.

[jmckitrick]

@vemv I found the source of the patch!

https://lists.gnu.org/archive/html/emacs-diffs/2022-05/msg00142.html

I was mistaken in thinking this was in emacs master. It's actually on a feature branch. I wonder where the discussion would happen, if this is slated to become part of emacs.

[vemv]

Nice!

I can't promise we'll give this a try super soon (other than a quick look and trial/error - already done).

But leaving the issue open seems reasonable, for visibilty. I hear that this capability would be benefitial for LSP performance. OTOH, there's more than one LSP offering for Emacs itself (eglot, lsp-mode), so it wouldn't surprise me if the existent ones are good-enough for the vast majority of cases.

Which is to say, sometimes it's wise to stick to what the authors wanted us to use.

Note that https://lists.gnu.org/archive/html/emacs-diffs/2022-05/msg00142.html is a mailing list item. You can certainly try bumping that conversation or perhaps contacting the author.

[jmckitrick]

@vemv I emailed him yesterday. I'll report back what I find out.

[jmckitrick]

Update: The original author of the feature is not able to work on it at this time, and I'm struggling to get any feedback at all on it.

I'm not sure if this issue is closer to the emacs and poll side, or the cider and nrepl side. I'm open to suggestions on how to proceed. Maybe figuring out where the timeout is actually occurring and if it's just a matter of adjusting how cider/nrepl connects?