@jespercockx I think you missed this commit.

Agda dev 2025-11-12: This PR should be tested by several users for the look-and-feel of the new behavior.
Volunteers:

• @nad
• @AndrasKovacs

Emacs complained when I tried to compile your code (make install-bin compile-emacs-mode):

In toplevel form:
[…]/Agda-2.9.0/emacs-mode/agda2-mode.el:2257:1:Error: the function ‘xref-apropos-regexp’ is not known to be defined.
Unable to compile the following Emacs Lisp files:
  […]/Agda-2.9.0/emacs-mode/agda2-mode.el

I'm using Emacs 27.1.

I have tried this PR with Emacs 30.2 and it seems to work fine. The basic jump-to-definition function seems to behave the same as before.

@nad Can you try upgrading xref from ELPA? I am not sure if the Makefile uses local packages when compiling. Otherwise I can look to redefine/backport the necessary definitions.

There's a conflict here now after merging some of your other PRs @phikal, could you please take a look?

Should be resolved.

I still cannot do a clean rebase, I could squash and merge but then that would defeat the purpose of splitting it in three commits.

I've now manually rebased the commit onto agda/master, lets see if that makes a difference.

I've now manually rebased the commit onto agda/master, lets see if that makes a difference.

Yes, that seems to have done the trick!

@nad did you have a chance to try out the suggestion of @phikal ?

@nad Can you try upgrading xref from ELPA?

I don't think we should require users of Agda to do that.

I don't think we should require users of Agda to do that.

While I don't think that it is much to impose (especially if users install Agda via their package manager, then that can also install the newer version of Xref for them as well; but I'd also expect that most people are using releases of Emacs newer than 27), I can look into back-porting the newer definitions if you think that is the right way to go ahead.

I can look into back-porting the newer definitions if you think that is the right way to go ahead.

I pushed a commit doing this. The code is directly from Emacs, but as it is does not constitute a significant contribution, it shouldn't pose a copyright issue.

@phikal the CI now gives an error on the elisp files:

In toplevel form:
.stack-work/install/x86_64-linux/1a3b758b945daa5d5a60949a822a32163a9ac421ebdac3111388f9237428b53b/9.12.2/share/x86_64-linux-ghc-9.12.2-9b33/Agda-2.9.0/emacs-mode/agda2-mode.el: Error: the function ‘apropos-parse-pattern’ is not known to be defined.
Unable to compile the following Emacs Lisp files:
  /home/runner/work/agda/agda/.stack-work/install/x86_64-linux/1a3b758b945daa5d5a60949a822a32163a9ac421ebdac3111388f9237428b53b/9.12.2/share/x86_64-linux-ghc-9.12.2-9b33/Agda-2.9.0/emacs-mode/agda2-mode.el
make: *** [Makefile:664: testing-emacs-mode] Error 1

Could you take a look please?

Looks like I just forgot to declare and require the right function, sorry about that!

@nad could you check again if this PR works for you now?

I'm pretty sure this broke the interactive highlight level for me, the highlighting does not go away. I'll keep testing and digging further

The CI failures appear to be unrelated to my change, right?

Either way, I have built a local Emacs 27 instance to check it myself, and while the byte-compiler doesn't complain, it appears I have some issue with resolving the Xref method... I cannot say if this is my mistake or not, so I'll have to take another look at this.

I have revised the PR onto master, throwing out the changes to annotation.el to make it less controversial (I recall that @nad was not happy with that) and change as little code as necessary. As the commit mostly just adds new code, the issues that @VojtechStep observed are also gone.

@phikal thank you for the update! Could you please fix the whitespace issue?

@nad are you happy with the expanation of @phikal or would you like to test it again on your system?

Fixed, was not aware that tabs were forbidden everywhere. BTW, has adding a EditorConfig ever been discussed?

Fixed, was not aware that tabs were forbidden everywhere. BTW, has adding a EditorConfig ever been discussed?

Not as far as I know, currently we just use fix-whitespace by @andreasabel but if we could replicate its functionality with a standard tool such as EditorConfig then I'd support that.

The main question would be whether we can also integrate it into CI.

I tested this PR now. I used the following code:

A : Set₁
A = Set

B : Set₁
B = A

If I load the file, go to the final occurrence of A, and type M-., then I get the following error message:

cl-no-applicable-method: No applicable method: xref-backend-definitions, agda2, "A"

Output from M-x emacs-version:

GNU Emacs 27.1 (build 1, x86_64-pc-linux-gnu, GTK+ Version 3.24.33, cairo version 1.16.0) of 2024-09-17, modified by Debian

I could reproduce that issue and debug it on a local Emacs 27 build. Turns out that in Emacs 27's cl-generic the specializer (eql 'agda2) is not the same as (eql agda2), which we should have used. Interestingly cl-typep doesn't work this way even on later versions? So I have adjusted that in my last force push and it works as intended (on my machine).

@nad do you want to test again or are you happy with @phikal 's explanation?

Is there a backwards compatibility policy for how many versions of emacs agda-mode ought to support? I know emacs 27 is what debian ships with, but it is nearly 5 years old at this point.

Is there a backwards compatibility policy for how many versions of emacs agda-mode ought to support? I know emacs 27 is what debian ships with, but it is nearly 5 years old at this point.

Debian stable has Emacs 30.1, Ubuntu LTS has 29.3, but according to this site there still are supported distributions like Rocky Linux 8 that have 26.1.

@nad do you want to test again or are you happy with @phikal 's explanation?

Now the code seems to work for me.