How to set formatting options for jdtls?

[parbo]

I've tried a bunch of different ways i found in old issues/discussions to set the java.format.settings.url and java.format.settings.profile settings, but I can't get it to work. How do I do it?

[parbo]

Maybe it doesn't work because I use eglot-java, it passes its own options: https://github.com/yveszoundi/eglot-java/blob/main/eglot-java.el#L146

[yveszoundi]

@parbo , eglot-java can now merge custom initialization options on top of defaults.

I added an example with the Eclipse Google formatting style.

[joaotavora]

I didn't know such a large package existed, but I'd just like to say it's not an official package, I can't recommend it. As far as I know everything needed to support LSP jdtls is in eglot.el. Have you read: https://joaotavora.github.io/eglot/#Advanced-server-configuration ? João

…

On Thu, May 11, 2023 at 9:22 AM Pär Bohrarper ***@***.***> wrote: Maybe it doesn't work because I use eglot-java, it passes its own options: https://github.com/yveszoundi/eglot-java/blob/main/eglot-java.el#L146 — Reply to this email directly, view it on GitHub <#1222 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAC6PQYRHI53FLSSA3IUBQTXFSOSTANCNFSM6AAAAAAX5H4OSM> . You are receiving this because you are subscribed to this thread.

[parbo]

I guess all the stuff I found in forums was outdated. After reading that I tried doing this:

  (add-to-list 'eglot-server-programs  `(java-mode . ("jdtls" :initializationOptions
  						     (:settings
  						      (:java
   						       (:format (:enabled t :settings (:url ,(expand-file-name (locate-user-emacs-file "cache/eclipse-java-google-style.xml"))
   											    :profile "GoogleStyle"))))
   						      :extendedClientCapabilities (:classFileContentsSupport t)))))

It seems to get sent correctly looking at the eglot logs, but jdtls still formats the same way.

[joaotavora]

Many servers don't document exactly the form of :initializationOptions they support, if any. I suggest you contact the server developers to know if and how :initializationOptions is supported. You can also try command-line options too to configure a server as suggested in "5.2 User-specific configuration" of the Eglot manual. What to pass there? Maybe https://github.com/eclipse/eclipse.jdt.ls#running-from-the-command-line holds clues. João

[joaotavora]

I feel that this is not an issue of failing to send the required initialization options to the language server.

That's a reasonable conclusion, from what I see there's nothing blatantly wrong. But sometimes some servers are very quirky and need information to be passed in very specific ways. For example I see two :settings objects in your JSON object. Is that exactly how it should look?

What would be a good way to debug why lsp-java mode can do this and eglot can't?

If lsp-java mode allows it, you could try to look at its logs to see exactly which JSON object(s) it sends instead of the one (or along the ones) that Eglot was sending.

[cayhorstmann]

Yes, it is two settings, as you can see from the TypeScript in the link that I provided. The first one is in InitializationOptions, the other in FormatOption.

Here is the section that is related to formatting in the lsp log:

Params: {
  "processId": 2373213,
  "rootPath": "/tmp/hello",
  ...
  "initializationOptions": {
    "settings": {
      "java": {
        ...
        "format": {
          "enabled": true,
          "settings": {
            "url": "/home/cay/bin/cay-eclipse.formatter.xml",
            "profile": "cay"
          },
          "comments": {
            "enabled": true
          },
          "onType": {
            "enabled": true
          },
          "tabSize": 3,
          "insertSpaces": true
        },
        ...
      },
      ...
  },
  ...
}

I now set all of these in eglot and see this in the logs:

(:jsonrpc "2.0" :id 1 :method "initialize" :params
          (:processId 2403949 :rootPath "/tmp/hello/" :rootUri "file:///tmp/hello" :initializationOptions
                      (:settings
                       (:java
                        (:format
                         (:enabled t :comments
                                   (:enabled t)
                                   :onType
                                   (:enabled t)
                                   :tabSize 3 :insertSpaces t :settings
                                   (:url "file:///data/cay/bin/cay-eclipse.formatter.xml" :profile "cay")))))
                                   ...

Here are the logs for invoking the formatter. In lsp-mode:

[Trace - 07:12:49 ] Sending request 'textDocument/formatting - (33)'.
Params: {
  "textDocument": {
    "uri": "file:///tmp/hello/NotHelloWorld.java"
  },
  "options": {
    "tabSize": 3,
    "insertSpaces": true,
    "trimTrailingWhitespace": true,
    "insertFinalNewline": true,
    "trimFinalNewlines": true
  }
}

And in eglot:

(:jsonrpc "2.0" :id 3 :method "textDocument/formatting" :params
          (:textDocument
           (:uri "file:///tmp/hello/NotHelloWorld.java")
           :options
           (:tabSize 8 :insertSpaces t :insertFinalNewline t :trimFinalNewlines t)))

Still, running eglot-format-buffer invokes the default formatter. Never mind the tab size; the brace style is completely different.

[joaotavora]

Ah I see. The problem is that those options you set before are not what influences eglot-format at all:

Eglot works in Emacs, and integrates with it. In Emacs, there are standard variables such as tab-width , to control tab size, indent-tabs-mode to control whether to insert tabs instead of spaces, and so on. In particular, the options for textDocument/formatting are calculated from these four variables:

:options (list :tabSize tab-width
                      :insertSpaces (if indent-tabs-mode :json-false t)
                      :insertFinalNewline (if require-final-newline t :json-false)
                      :trimFinalNewlines (if delete-trailing-lines t :json-false))

So, set the variables tab-width, indent-tabs-mode ,require-final-newline and delete-trailing-lines and try again.

[cayhorstmann]

Like I said, never mind the tab width. I know how to fix that later. The problem is the brace style. I want jdtls to use the particular formatting rules in that xml file, which is something that it is able to do via lsp-java. But not eglot.

[joaotavora]

From the information you've given me, the only discrepancy I see in LSP communication was caused by the 4 inconsistent Emacs variables I noted. From what I can detect in a comparative analysis and from the information you've given me everything else is the same.

If you'd like to create a proper complete bug report with all the elements for a reproduction of this issue, we might be able to help you further. Please follow the guide at https://joaotavora.github.io/eglot/#Troubleshooting-Eglot

[zsxh]

I think this is related to #1197 (reply in thread)

eglot sends empty didChangeConfiguration to jdtls.

[joaotavora]

On Thu, May 11, 2023 at 12:53 PM zsxh ***@***.***> wrote: I think this is related to #1197 (reply in thread) eglot sends empty didChangeConfiguration to jdtls.

I don't see the relation, can you clarify? Also didChangeConfiguration is for project-specific server configuration. Here there is no want for project specific server configuration -- as far as I can tell -- rather user-specific configuration The manual also explains how to to do project-specific server configuration. João

[zsxh]

Yes, I confirm that it's related to didChangeConfiguration.

(package-initialize)
(global-company-mode)
(yas-global-mode)
(require 'eglot)
(add-to-list
 'eglot-server-programs
 `(java-mode . ("jdtls"
                :initializationOptions
                (:settings
                 (:java
                  (:completion (:guessMethodArguments t)))
                 :extendedClientCapabilities (:classFileContentsSupport t)))))

/tmp/demo
├── Abc.java
└── .git

import java.util.ArrayList;
import java.util.List;

/**
 * Abc
 */
public class Abc {

    public static void main(String[] args) {
	String txt = "txt";
	List<String> alist = new ArrayList<>();
        alist // type dot to trigger completion
  }
}

open Abc.java, M-x eglot, you can see logs like this:

...
(:jsonrpc "2.0" :method "workspace/didChangeConfiguration" :params
	        (:settings #s(hash-table size 1 test eql rehash-size 1.5 rehash-threshold 0.8125 data
				                           ())))

alist trigger completion

When I prevent sending empty didChangeConfiguration, config works fine to me.

;; NOTE: hack workspace/didChangeConfiguration
(define-advice eglot-signal-didChangeConfiguration (:override (server) override)
    "Send a `:workspace/didChangeConfiguration' signal to SERVER.
When called interactively, use the currently active server"
    (interactive (list (eglot--current-server-or-lose)))
    (when-let ((settings (eglot--workspace-configuration-plist server)))
      (jsonrpc-notify
       server :workspace/didChangeConfiguration
       (list :settings settings))))

[skyler544]

Thank you, this somehow seems to enable the use of the formatting options as well. I tested it on GNU Emacs 29.1.90 (build 1, x86_64-pc-linux-gnu, GTK+ Version 3.24.38, cairo version 1.18.0) of 2023-10-26 with this procedure:

1. Run $ emacs -q
2. Evaluate following in the scratch buffer:

(require 'eglot)
(add-to-list 'eglot-server-programs
             '(java-mode . ("jdtls" :initializationOptions
                            (:settings
                             (:java
                              (:format
                               (:enabled "true"
                                         :settings
                                         (:url "path-to-your-style.xml"))))))))

3. Open a java file that is not formatted according to your style file (notice the very long lines):

package org.comments.services;

import org.comments.proxies.CommentNotificationProxy;
import org.comments.repositories.CommentRepository;

public class CommentService {
    private final CommentRepository commentRepository;
    private final CommentNotificationProxy commentNotificationProxy;

    public CommentService(CommentRepository commentRepository, CommentNotificationProxy commentNotificationProxy) {
        this.commentRepository = commentRepository;
        this.commentNotificationProxy = commentNotificationProxy;
    }
}

4. M-x eglot
5. M-x eglot-format-buffer
6. Nothing happens
7. Go back to step 1 and start over, this time including the code provided by zsxh.
8. In step 6 the buffer will now be formatted, including line length limits and so on.

[cayhorstmann]

Do you mean the line (remove-hook 'eglot-connect-hook 'eglot-signal-didChangeConfiguration)? With that line added, I too get the custom formatter to work.

Is that a safe fix, or just something to pinpoint the issue?

[joaotavora]

Do you mean the line (remove-hook 'eglot-connect-hook 'eglot-signal-didChangeConfiguration)? With that line added, I too get the custom formatter to work.

Aha! So it means that jdtls forgets previous command line options when workspace-specific options are also given. Please @cayhorstmann comment this in the Emacs issue you created.

[joaotavora]

On Thu, May 11, 2023 at 1:50 PM zsxh ***@***.***> wrote: Yes, I confirm that it's related to didChangeConfiguration.

No, there's absolutely no relation to the link comment you posted. It would be the same as saying my washing machine's manual and Mozart's opera are related because they're both written in German. :initializationOptions is NOT the same as "workspace configuration". Please read the manual link I posted. I just know it is possible to configure servers in two fundamentally different ways, project-specific and user-specific, as described in the manual. It shouldn't -- in theory -- need any hacks. If user-specific configuration that this OP is trying to achieve is not supported by jdtls server and only project-specific configuration is absolutely required for the jdtls server in particular -- that's for the jdtls devs to clarify. If that's the case then, again, the manual has information of how to do that. João

[joaotavora]

Aha! So it means that jdtls forgets previous command line options when workspace-specific options are also given.

@zsxh I'm sorry if that's what you meant by the relation between user-specific and project-specific configuration.
If it's true, it's a server-specific relation, but indeed a relation.

[skyler544]

Do you mean the line (remove-hook 'eglot-connect-hook 'eglot-signal-didChangeConfiguration)?

I didn't know about this fix; I meant all of this:

  (define-advice eglot-signal-didChangeConfiguration (:override (server) override)
    "Send a `:workspace/didChangeConfiguration' signal to SERVER.
When called interactively, use the currently active server"
    (interactive (list (eglot--current-server-or-lose)))
    (when-let ((settings (eglot--workspace-configuration-plist server)))
      (jsonrpc-notify
       server :workspace/didChangeConfiguration
       (list :settings settings))))

I've had this enabled since I commented and it hasn't caused any issues. I just tried your suggestion of (remove-hook 'eglot-connect-hook 'eglot-signal-didChangeConfiguration) and it also seems to enable the custom formatting. I'll test it for a few days and see if anything weird happens.

Caveat lector I haven't looked into any of this deeply enough to see why these configuration snippets would or wouldn't work, I was just looking through the issue tracker and these discussions trying to see if I could find hints on configuring the jdtls formatter settings and found this thread.

[cayhorstmann]

Hi, that syntax worked for me, thanks! I did need the :profile "cay" because my formatter has multiple profiles. I also managed to add in runtime configurations:

((java-mode
  . ((eglot-workspace-configuration
      .
      (:java
       (:format
        (:enabled t 
                  :settings
                  (:url "file:///data/cay/bin/cay-eclipse.formatter.xml"
                        :profile "cay")))
        :configuration
        (:runtimes
         [(:name "JavaSE-11" :path "/data/apps/jdk-11")
          (:name "JavaSE-17" :path "/data/apps/jdk-17")
          (:name "JavaSE-21" :path "/data/apps/jdk-21" :default t)]))
      ))
  ))

Now what would be the best way of achieving this globally, without a per-project file?

[joaotavora]

Now what would be the best way of achieving this globally, without a per-project file?

It's in the manual. Use setq-default. But the right way is for jtdls to address the fact that empty workspace configuration completely overrides, instead of merging, initializationOptions.

Another possibly more correct approach is keep doing the initializationOptions (like here and add this:

(add-hook 'java-mode-hook (lambda ()
   (remove-hook 'eglot-connect-hook #'eglot-signal-didChangeConfiguration t))

So that -- in java-mode only -- Eglot refrains from sending workspace options.

[cayhorstmann]

I did that, and I added the runtime configurations like this:

(add-to-list 'eglot-server-programs
             `(java-mode . ("jdtls"
                            :initializationOptions
                            (:settings
                             (:java
                              (:format
                               (:enabled t 
                                         :settings
                                         (:url "file:///data/cay/bin/cay-eclipse.formatter.xml"
                                               :profile "cay")))
                              :configuration
                              (:runtimes
                               [(:name "JavaSE-11" :path "/data/apps/jdk-11")
                                (:name "JavaSE-17" :path "/data/apps/jdk-17")
                                (:name "JavaSE-21" :path "/data/apps/jdk-21" :default t)]))
                             ))
                         ))

The outcome: jdtls uses my custom formatter. jdtls does not use the default runtime.

Why does act differently from using .dir-locals.el?

[joaotavora]

I did that

What is "that"? I suggested two approaches in my post: one to globally set eglot-workspace-configuration, another one to nullify it only in java-mode.

Why does act differently from using .dir-locals.el?

initializationOptions is not the same as eglot-workspace-configuration. They're two different LSP interfaces, and the server does with them what it wants to do. If you find this behavior incongruent, talk to the server devs.

Here in the Eglot forum, we're mostly interested in things that Eglot sends to the server (if it's sending them in the correct formats and at the correct time) or incorrect interpretations of things Eglot receives from the server.

When I say "mostly", it doesn't mean we're exclusively interested in that. It's fine to chat about server things. But, as much as it is easy to to regard Eglot as an abstraction umbrella over all things LSP in Emacs, it is not meant to abstract away the servers. LSP's stated goal is to do that, not Eglot's. In Eglot -- perhaps contrary to lsp-mode -- the user must be minimally aware of what server they are running. Some servers keep the needed brainpower to a comfortable minimum, some don't. You picked one of the latter 😄 (though I understand you didn't have much choice).

[yveszoundi]

@cayhorstmann, I'm no Eclipse JDT LS expert, but you could try referencing the default runtime as JAVA_HOME inside the :java block. Add a new line prior the (:format ....) block, with the following contents:
(:home "/data/apps/jdk-21")

Not sure if the above works, but it might be worth a try. As @joaotavora mentioned, the Eclipse JDT LS server is a bit involved, and they might be able to answer better specific questions.

[HaraldKi]

As I went through all that trouble and learned a lot from this thread and others, let me try to summarize my solution:

1. Remove eglot-signal-didChangeConfiguration from eglot-connect-hook as described above. Not doing so will end you up in the madhouse, as there is no way to understand why nothing works. I opened a bug for eclipse.jdt.ls.

2. These initializationOptions work for me.

(add-to-list
   'eglot-server-programs
   `((java-mode hk-java-mode)
     .
     ("jdtls"
      :initializationOptions
      (:settings
       (:java
        (:format (:enabled t :profile "harald-variant" :settings (:url "/home/harald/.emacs.d/eclipse-formatter.xml"))
         :settings (:url "/home/harald/.emacs.d/eclipse-settings.prefs")
         ))))))

The details of how to structure the lispified JSON are documented, though it is no doubt a painful to piece it together from all those interfaces, not the least because a :settings field may appear in at least three (sub)interfaces.

For better control I also did this:

3. Download a recent version of eclipse.jdt.ls. Don't get confused by the non-numerical sorting when picking the most recent one. 1.33 works for Java-21, 1.34 is for Java-22 which gets really important when using preview features.

4. Use a script to best control how jdtls is started, something like:

#!/bin/bash
# points to my current used version of jdt.ls
JDTLS=$(readlink -f ~/tools/Store/jdt-language-server)
VERSION=${JDTLS#*-server-}
# set it explicitly to be sure you know where this all ends up
DATA=~/work/jdtls-data
#DEBUG="-agentlib:jdwp=transport=dt_socket,address=127.0.0.1:8888,server=y,suspend=n"
#TRACE=-Djdt.ls.debug=true
cd "$DATA"
exec -a "jdtls($VERSION): java" java \
	-Declipse.application=org.eclipse.jdt.ls.core.id1 \
	-Dosgi.bundles.defaultStartLevel=4 \
	-Declipse.product=org.eclipse.jdt.ls.core.product \
	-Dlog.level=ALL \
        $DEBUG $TRACE \
	-Xmx1G \
	--add-modules=ALL-SYSTEM \
	--add-opens java.base/java.util=ALL-UNNAMED \
	--add-opens java.base/java.lang=ALL-UNNAMED \
	-jar "$JDTLS/plugins/org.eclipse.equinox.launcher_"*.jar \
	-configuration "$JDTLS/config_linux" \
	-data "$DATA"

The script assumes that you have set up the JDK you want to use in your PATH. In this case, setting JAVA_HOME should never be necessary. Rather check and make sure it is not set.

5. When changing the jdt.ls version or when things start to behave weird, stop the jdt.ls server(s) and clear your $DATA directory. I had very annoying behavior sometimes and clearing $DATA usually brought back some sanity.

6. Check $DATA/.metadata/.log for hints if things go wrong, though it does not contain a lot.