Support for multiple connections

[technomancy]

It seems like right now monroe only supports connecting to a single nrepl instance at a time.

Are you interested in adding support for more than one active *monroe* buffer, or is it better to use separate Emacs instances for that?

[sanel]

I was planning to add it (made a few attempts), just didn't have enough time to complete it. However, if you are able to do it, keeping the code simple, I'll be happy to accept the patch :)

[technomancy]

Good to know; thanks. No promises, but I might take a shot at this at some point.

[bbuccianti]

I'm thinking of making an attempt too. Maybe we can discuss a little about how to do it?

[sanel]

Feel free @bbuccianti :) Maybe check inferior-lisp mode (inf-lisp.el in Emacs) for multiple processes support.

[sogaiu]

@bbuccianti

For reference, if you want to see an existing attempt for a socket repl-based thing with multiple connections, I've had some luck w/:

• https://github.com/sogaiu/acrepl/blob/arcadia-and-reconnect/acrepl-connect.el
• https://github.com/sogaiu/acrepl/blob/arcadia-and-reconnect/acrepl.el

It's been working pretty well for me so far.

I started with sesman, but decided it was much more than I needed.

[bbuccianti]

@sanel @technomancy @sogaiu

Can you help me test this?

https://github.com/bbuccianti/monroe/

I'm sure there is a lot of ways to improve it, please help me!

[sogaiu]

Do you mean the multiple-connections branch?

https://github.com/bbuccianti/monroe/tree/multiple-connections

[bbuccianti]

This is the correct commit:

bbuccianti@38a1488

[sogaiu]

Had success interacting with 2 separate processes from within a single emacs instance.

Opened two files -- each one from a different project -- ran M-x monroe in each and tried M-x monroe-eval-region. Seems to work 👍

IIUC, there is at most one connection per project and at most one process per project. Is that correct?

[bbuccianti]

Yes, it's using the project folder as a pointer concatenating it with monroe and monroe-conn buffers.

I've created a pull request #33

[sanel]

Thanks @bbuccianti for your work and @sogaiu for fast response!

Is #33 will add support for multiple connections? Title of PR confuses me :)

Also, does it handle reading .nrepl-port from project tree?

[bbuccianti]

It still uses monroe-locate-port-file.

The only change is that instead of naming the buffers *monroe* and *monroe-conn* I conj the project folder to it. Then, when you try to send code to the nrepl it searches for the correct buffers using the project folder.

[bbuccianti]

@sanel I just improved the title of the PR

[sanel]

Thanks! Here are few issues I found (my Emacs is 26.3). After first expression is entered (in single or multiple REPL connections), prompt is not shown any more. Next issue is connecting to REPLs outside of working directories. For example, you run REPL at `localhost:7888` in project `foo`, then in external terminal run REPL in project `baz` at `localhost:7889`, but keep Emacs in project `foo`, it will connect to `foo` for both REPL and inputs and outputs will be shared. The same applies if you connect Emacs to any multiple projects without entering to project folders. Since you are calculating process name from working directory, I'd suggest to use combination of host+port instead, because they have to be unique on given system. Also, I'd suggest to pop out exising monroe connection REPL if user attempt to connect to that session again. This will prevent multiple REPLs that will echo inputs and outputs between each other.

[bbuccianti]

I can't reproduce the issue with prompt. Maybe I'm evaluating trivial things, but seems to work fine for me.

bbuccianti@05b37d9

Now it's using the host:port combination to name buffers and also call (monroe-disconnect) before trying to attempt another connection.

Let me know!

[sanel]

Thanks @bbuccianti for your work so far! However, I still have issue with prompt. What Emacs are you using?

Here is how to reproduce it:

1. Run lein repl somewhere.
2. In new Emacs session, run monroe, it will ask for connection.
3. After you connect, you will see prompt. Enter something like (+ 1 2 3) in REPL. It will print result, but prompt is no longer visible.

Here is how it looks for me:

user=> (+ 1 2 3)                                                                                                                                                                                                    
6
;; no prompt here                                                                                                                                                                                                                   
(println "abc")                                                                                                                                                                                                     
abc
;; no prompt here

[bbuccianti]

Are you using this? #33

I'm currently using GNU Emacs 26.3 (build 1)

[sanel]

Yes, using PR #33 and same Emacs version. Tried even with emacs -Q.

Can you try loading PR with emacs -Q please?

[bbuccianti]

Well, that works for reproducing the issue. Any suggestion on how to fix it?

[sanel]

I'm curious how it worked for you - maybe some package "fixed" that :) This happened before for me and I believe it is up to comint-mode now as well (comint can break in weird ways if you messed up options or processes). I'll check this one.

[bbuccianti]

I've tested with commit 2f472fd and it's not working in emacs -q !

[sanel]

Hm... does it work for you if you revert to this: a7cf37e ?

[bbuccianti]

Yeah! Can confirm that commit a7cf37e it's working as expected.

Also I want to tell that seems like a callback done with monroe-make-response-handler isn't running with the "done" status. Then monroe-requests isn't cleaned so the prompt is never added again.

But is really strange, because status is clearly truthy but can't go inside the (when status ;cleaning) code. I'm really new to those kind of callbacks and don't know really how monroe-dbind-response is doing is magic, but I'm learning a lot. Would be great if you @sanel can give some advice in how to debug that part.

Thanks!

[bbuccianti]

@sanel no advances on this?

[sanel]

Hey Benjamín, sadly I'm a bit busy currently :S If you have time to take a look at it, I'll be happy to review it.

[bbuccianti]

I force-pushed to #33

Thanks @sanel

[sanel]

Thanks @bbuccianti ! Will take a look :)

[sanel]

Implemented with #33

[sanel]

Sorry @bbuccianti but this still doesn't work. To demonstrate:

• Emacs 26.3
• In separate run new nrepl server with lein repl
• From emacs, run monroe and connect to nrepl address
• In monroe buffer, enter + and press enter
• It will print #object[clojure.core$_PLUS_ 0x4e0e22f4 "clojure.core$_PLUS_@4e0e22f4"] but prompt is not shown.

Worse, with multiple connections, it mess up where evaluation should go. To reproduce:

• Run 2 different nrepl servers it separate terminals.
• Connect from Emacs to both of them (running monroe and entering their addresses).
• Enter + in one monroe buffer, it will evaluate in second.

How you manage to get this working?

[sanel]

@technomancy after 1c2559b prompt is no longer shown when you enter something like + in fresh monroe session. What is reasoning behind this change?

Does it work for you?

[bbuccianti]

I think it's my error. I assumed that you would open a monroe session from a project folder. So it's searching for the .nrepl-port file in order to get the correct buffer to send eval string to the correct process.

That's why connecting from another places it's screwing everything. I don't know how to solve that problem (yet), I need to think more.

Maybe we can add a prompt in order to ask for the project folder when it doesn't find that .nrepl-port file?

[bbuccianti]

I'm testing b540e13 in Emacs 28.0.5 and seems to resolve ok different connections, without the need of .nrepl.-port. I read now that it tries to get port from the .nrepl-port file, if it doesn't find one it is extracted from the buffer. That's way the buffer now it's called "monroe: %s" being %s the host:port.

Maybe it's an issue with Emacs 26.3 version? But I think @technomancy was using that version of 25, I don't remember exactly.

[sanel]

I had the same issue with 26.1 I think... we must support older versions as well, including currently stable one :)

Now, I haven't dig deeper into the code, but the strategy of reading ports should be different IMO. When new connection is requested, it should only read .nrepl-port file to suggest host and port (as was doing before). After that, it should encode those inside buffer name.

I'm curious, how clojure-mode buffer (buffer where clojure code is) knows to which monroe repl to send code?

[technomancy]

@sanel re: 1c2559b the old logic was only to print the prompt when a message was received that did not contain certain keys in it; however, IMO this was taking advantage of an implementation detail in the reference implementation of the Clojure nrepl server.

There is nothing inherent in "a message with no output" that guarantees it is equal to "the input you sent is done being evaluated"; it's just an accidental property that the one nrepl server happens to have, and other servers do not have. In that commit I made it based on whether there were any outstanding requests in the monroe-requests table, but that's no longer a good criteria now that multiple connections are possible.

Probably what it should have been based on from the start is when a message is received where the status field contains "done".

[bbuccianti]

It's still the same approach. It reads from .nrepl-port or you can provide host:port or :port using localhost as default. Nothing changes there.

What changes is the name of the monroe buffer. Before it was monroe. Now with this change we create a buffer as monroe: host:port. And for the connection it's similar. It creates a monroe-connection: host:port.

In this way then, to return the correct buffer we use monroe-repl-buffer function. If you found .nrepl-host, it returns the monroe: host:port buffer, otherwise it's assuming you are in the monroe buffer and it extracts host:port from the buffer name.

[bbuccianti]

I see an error that I can't fix. I can show you the output of both versions of Emacs 26 and 27, reactioning to the same input.

decoded: ((dict (id . 1) (ns . user) (session . a54cd27f-152b-4b2e-aa87-384597d4486b) (value . 5)) (dict (id . 1) (session . a54cd27f-152b-4b2e-aa87-384597d4486b) (status done)))
net-filter response: (dict (id . 1) (ns . user) (session . a54cd27f-152b-4b2e-aa87-384597d4486b) (value . 5))
response: (dict (id . 1) (ns . user) (session . a54cd27f-152b-4b2e-aa87-384597d4486b) (value . 5))
net-filter response: (dict (id . 1) (session . a54cd27f-152b-4b2e-aa87-384597d4486b) (status done))
response: (dict (id . 1) (session . a54cd27f-152b-4b2e-aa87-384597d4486b) (status done))
GNU Emacs 27.0.91 (build 1, x86_64-pc-linux-musl, X toolkit, cairo version 1.16.0) of 2020-07-03

Monroe connected.
decoded: ((dict (id . 1) (ns . user) (session . fec70c4b-59a9-4258-a884-c6e270720ae3) (value . 5)) (dict (id . 1) (session . fec70c4b-59a9-4258-a884-c6e270720ae3) (status done)))
net-filter response: (dict (id . 1) (ns . user) (session . fec70c4b-59a9-4258-a884-c6e270720ae3) (value . 5))
response: (dict (id . 1) (ns . user) (session . fec70c4b-59a9-4258-a884-c6e270720ae3) (value . 5))
GNU Emacs 26.3 (build 1, x86_64-unknown-linux-musl) of 2019-08-29

I'm trying to investigate why net-filter isn't called again.

[bbuccianti]

I think I have a fix in #34